 |
|
Housing Vacancies and Homeownership (CPS/HVS)
|
|
|
|
Annual Statistics: 2007
APPENDIX B. SOURCE AND ACCURACY OF ESTIMATES
Instrument Changes in 2007
Beginning in January 2007, the CPS/HVS began using Blaise, a powerful
computer-assisted interviewing (CAI) system and survey processing tool for the Windows
operating system. It is being used for many of the surveys now being conducted by the Census
Bureau.
Methodology Changes in 2003
Population controls that reflect the results of the 2000 decennial census are used in the
CPS/HVS estimation process for the first time in the first quarter 2003. This change has a slight
effect on vacancy and homeownership rates, as described below. As a final additional step in
our estimation process, the estimates are now controlled to independent housing counts used for
the first time in order to produce a more accurate estimate of housing units. This should make
the CPS/HVS estimates of housing units more consistent with other Census Bureau
housing surveys. The new housing controls will affect the count of all housing units in the sense
that both occupied and vacant units will be ratio estimated to the new control total. Vacancy
rates and homeownership rates are not be affected by this change.
The CPS/HVS began computing first-stage factors (used for weighting purposes) based on
year-round and seasonal counts of housing units from the 2000 decennial census, beginning in
the first quarter 2003. From 1980 to 2002, the CPS/HVS first-stage factors were based on
year-round estimates only. We believe that this improves our counts of year-round and seasonal
units. The shift from 1990-based to 2000-based population controls (including the weighting
revision) had a very slight effect on vacancy rates and homeownership rates. Research has
shown that the new 2000-based controls dropped the rental vacancy rate in the first quarter 2002
from 9.14 percent to 9.08 percent---a difference of less than 1/10 of one percent. The
homeowner vacancy rate was revised from 1.67 percent to 1.65 percent, while the
homeownership rate was revised from 67.82 percent to 67.81 percent. The questions on race on
the CPS are modified beginning in the first quarter 2003 to comply with the revised standards for
federal statistical agencies. Respondents may now select more than one race. The
Hispanic/Nonhispanic origin question continues to be asked separately.
Introduction of Computed-Assisted Survey Collection (CASIC) and 1990
census-based weighting (from 1995 to 2002)
Major changes related to the Current Population Survey/Housing Vacancy Survey
(CPS/HVS) were effective beginning with the first quarter 1994 data. First, a new weighting
procedure was implemented based on the 1990 decennial census. The 1990-based weighting
produces, on average, estimates of the total housing inventory that are about 0.1 percent lower
than the 1980-based weighting. Revised data are provided in the historical tables for 1993 to
show the effect of this change. Generally, the vacancy rates are only minimally affected, while
the homeownership rate is about one-half of a percentage point lower with the new weighting
procedures.
A second change is that the CPS/HVS has become a totally computerized survey with the
implementation of the Computer Assisted Survey Information Collection (CASIC). The CASIC
tools consist of state-of-the-art computer-assisted modules for data collection and processing.
Although the concepts, definitions, and questionnaire items remain the same, the shift to CASIC
may affect vacancy rates and homeownership rates. We are unable to determine the quantitative
effects of the use of CASIC on the vacancy and homeownership rates. Data users should use
caution when comparing 1994 and later data with earlier data. Beginning the second quarter of
1999, a change was made in the way data for housing units in structure are collected. In the past,
there was one category to show 1-unit in structure. Now that has been separated into two
categories: 1-unit attached and 1-unit detached.
Source of Data
The estimates presented in this report are based on data obtained from two surveys
conducted by the Bureau of the Census. Data concerning vacancy rates and tenure of occupied
housing units are from the monthly sample of the Current Population Survey/Housing Vacancy
Survey (CPS/HVS). Characteristics of occupied housing units are from the American Housing
Survey (AHS).
CPS and AHS Designs
Since the inception of the CPS in 1940, the sample has been redesigned several times to upgrade
the quality and reliability of the data and to meet changing data needs. From July 1995 to March
2004, the CPS/HVS sample is selected from a frame based on the 1990 decennial census and is
spread over 754 sample areas, which represent 2,007 geographic areas in the United States.
Beginning in April 2004, the new sample drawn from the 2000 decennial census is phased in
over a 15-month period. From April 2004 to June 2005, the sample consisted of sample units
drawn from both the 1990 and 2000 decennial censuses. The metropolitan/nonmetropolitan data
shown for 2005 report reflect 2000 census definitions, while data shown for 1995 to 2004 reflect
1990 metropolitan/nonmetropolitan definintions.
Beginning in the first quarter 1986, vacant seasonal mobile homes were included in the count of
vacant seasonal units. This change resulted in a 12 percent increase in the number of vacant
seasonal housing units.
Beginning in 2002, the size of the CPS/HVS sample increased to approximately 72,000 housing
units. This expansion was one of the Census Bureau's plans to meet the requirements of the State
Children's Health Insurance Program (SCHIP) legislation. Of the 72,000 housing units contained
in the CPS/HVS sample, approximately 61,200 are eligible for interview each month; of this
number, 3,900 occupied units, on the average, are visited but interviews are not obtained because
occupants are not found at home after repeated calls or are unavailable for some other reason. In
addition to the 61,200, there are also about 10,800 sample units in an average month which are
visited but are found to be vacant or otherwise not to be interviewed. About half of the 10,800
are vacant and interviewed for the HVS. The CPS estimation procedure for occupied units
involves the inflation of the weighted sample results to independent estimates of the total civilian
noninstitutional population of the United States by age, race, sex, and Hispanic/non-Hispanic
categories. These independent estimates are based on statistics from the decennial censuses of
population; statistics on births, deaths, immigration, and emigration; and statistics on the strength
of the Armed Forces. The HVS estimation procedure for vacant units is similar to that used for
occupied units. Weighted sample results are adjusted at the state level using 2000 census vacant
counts. A second adjustment inflates these results based on the CPS coverage of occupied units
by geographic areas. As a final step for both the CPS and HVS, all housing unit counts are
adjusted to reflect independent housing control totals. This change is effective, beginning 2003.
Data shown in all tables (except table 2) on vacancy rates and tenure of occupied units are based
on a 12-month average for 2007. The data concerning the distribution of characteristics for
occupied housing units, shown in table 2, are obtained primarily from the AHS national sample.
Distributions of characteristics of occupied housing units from the AHS estimates are applied to
CPS current housing inventory independent estimates to obtain the characteristics of occupied
housing units used in this report. The Survey of Construction (SOC) and the Consumer Price
Index also are used to improve estimates of the rent distribution. The 2005 AHS sample is spread
over 394 sample areas comprising 878 counties and independent cities with coverage in each of
the 50 States and the District of Columbia. Of the 59,400 housing units both occupied and vacant
contained in the AHS sample, 56,650 were interviewed and 6,150 were classified as "Type A
noninterviews" for various reasons. 2,800 units were visited but were not eligible to be
interviewed for the purposes of AHS. A detailed description of the AHS sample design and
estimation procedure can be found in the H-150 report for 2005.
Comparability with Census of Housing Data
Most of the concepts and definitions are the same for items that appear in both the 1980
and 1990 censuses and the Housing Vacancy Survey. However, there is one minor difference in
the housing unit definition between the CPS/HVS and the 1980 and 1990 decennial censuses.
The difference is that, in the CPS/HVS, living arrangements containing five or more persons, not
related to the person in charge, were classified as group quarters; for the 1980 and 1990 census,
the requirement was raised to nine or more persons not related to the person in charge. There
were some differences in what has been counted as housing units between the earlier censuses
and the CPS/HVS. Descriptions of the differences between earlier censuses and the CPS/HVS
appear in the 1985 and earlier reports of this series.
Prior to 1990, there were significant differences between the CPS/HVS and the decennial
censuses. The 1980 and 1990 decennial censuses included vacant mobile homes as housing
units, whereas prior to 1986 the CPS/HVS did not. However, beginning in 1986, vacant seasonal
mobile homes were counted as housing units in the CPS/HVS. In addition, year-round vacant
mobile homes were counted as housing units, beginning in 1990 in the CPS/HVS. Another
difference in the housing unit definition between the CPS/HVS (prior to 1986) and the 1980 and
1990 censuses was that the CPS/HVS required units to be separate living quarters and have direct
access or have complete kitchen facilities. For the 1980 and 1990 decennial censuses, the
complete kitchen facilities alternative was dropped with direct access required of all units.
However, beginning in 1990, the CPS/HVS requirement for complete kitchen facilities was
dropped with direct access required of all units. Thus, the earlier definitional differences were
eliminated.
In addition, there are differences between the methodologies used to collect data for the
CPS/HVS and the censuses. These differences include interviewing procedures, staff experience
and training; differences in processing procedures and sample designs; the sampling variability
associated with the CPS/HVS and the sample data from the census; and the nonsampling errors
associated with the CPS/HVS and census data.
Research has shown that the CPS/HVS and the 1990 census produced significant
differences for vacancy characteristics. The rental vacancy rate from the April 1990 census was
8.5 percent, whereas, the CPS/HVS reported the rental vacancy rate of 7.2 percent for the first
half of 1990. The April 1990 census had a homeowner vacancy rate of 2.1 percent, while the
CPS/HVS had a vacancy rate of approximately 1.7 percent for the first half of 1990. For
occupied housing, the April 1990 census produced a homeownership rate of 64.2 percent, while
for the first half of 1990 the CPS/HVS produced a rate of 63.9 percent. These differences
illustrate that, for these characteristics as well as others, caution should be used when making
comparisons between the 1990 census and the CPS/HVS.
Further research has shown that the CPS/HVS and the 2000 decennial census produced
significant differences for vacancy characteristics. The rental vacancy rate from the April 2000
census was 6.8 percent, whereas the CPS/HVS reported the rental vacancy rate of 7.9 percent for
the first half of 2000. The April 2000 census has a homeowner vacancy rate of 1.7 percent for
the first half of 2000. For occupied housing, the April 2000 census produced a homeownership
rate of 66.2 percent, while for the first half of 2000, the CPS/HVS produced a rate of 67.2
percent. These differences illustrate that, for these characteristics as well as others, caution
should be used when making comparisons between the 2000 census and the CPS/HVS.
Comparability with Earlier Data
As stated earlier in this report, beginning in 1994 new weighting procedures based on the
1990 decennial census were implemented. In addition, the survey data collection procedures
became totally computerized. Caution should be used when comparing current data with
unrevised data prior to 1994.
In 1989, new edit procedures were implemented in the Current Population
Survey/Housing Vacancy Survey (CPS/HVS). These new procedures were used to allocate cases
that would have been classified as "not reported" under previous procedures.
In 1990, year-round vacant mobile homes were included for the first time as part of the
year-round vacant count of housing units. This change was made to make the composition of the
housing unit inventory for the CPS/HVS similar to the decennial census and other surveys, which
count all mobile homes as housing units when occupied or vacant (available for occupancy on
the site). Research has shown that the inclusion of year-round vacant mobile homes increases the
vacancy rate significantly in some cases. All of the 1989 data in this report have been updated to
include year-round vacant mobile homes. Caution should be used when comparing unrevised
vacancy data prior to 1990 to data for later years.
In addition to the above mentioned design and estimation changes, caution should be used
in comparing data for 1980 and beyond in this report with data from 1979 and earlier years.
Starting in 1980, several changes were implemented in the survey to improve the reliability of the
data presented. These included adding a supplemental sample, refining the estimation
procedures, and changing the source of occupied characteristics from the Quarterly Housing
Survey (QHS) to the AHS.
Although the above mentioned changes have resulted in more reliable estimates, data for
1980 and later in this report are not completely comparable to data for 1979 and previous years, as
published in Housing Vacancies reports, series H-111. Furthermore, unrevised data prior to 1990
are not completely comparable to 1990 data and beyond, due to the inclusion of year-round vacant
mobile homes, beginning in 1990. Thus, particular caution should be observed in drawing
conclusions about trends that extend from before 1980 to 1980 and beyond, and also trends from
before 1990 to 1990 and later. For comparative purposes, 1979 data in this report have been
revised to incorporate all changes made in 1980, and 1989 data have been revised to incorporate
all changes made in 1990. Unrevised 1989 and 1979 data are provided to show the magnitude of
the various changes.
The revised 1979 vacancy estimates are higher than the original 1979 estimates. The
increase in vacancy rates was not the result of locating additional vacant units, but reflects the
increase in sample size and refinements in the estimation procedure. It is safe to assume that prior
to the implementation of these new procedures (1955 through 1978) HVS produced
underestimates of vacant units. Earlier reports in this series give more complete descriptions of
the original CPS sample, the QHS sample, and estimation procedures.
Caution in Using Vacancy Rates for Characteristics in Table 2
Vacancy rates in table 2 are based in part on forecasts of occupied housing units. These
forecasts are periodically revised to incorporate more recent data and improved forecasting
procedures. Data shown for 2007 and 2006, shown on table 2 are based on the 2005 AHS.
For the occupied unit forecasts for the monthly rent categories, we update the AHS data
quarterly to reflect the rise in the cost of renting through the use of the residential rent index, and
the latest available asking rent data for newly constructed rental units.
Caution in Using Seasonal Vacant Data
Analysis of seasonal vacant data prior to the 1987 has shown that estimates for these
characteristics have been underestimated by approximately 28 percent. The estimates beginning
in 1987 are adjusted to reflect this. This revision has an effect on other categories (especially the
percentage occupied) in addition to seasonal vacant units in the distributions shown in tables 7.
Accuracy of the Estimates
Since the CPS/HVS estimates are based on a sample, they may differ somewhat from the
figures that would have been obtained if a complete census had been taken using the same
questionnaires, instructions, and enumerators. There are two types of errors possible in an
estimate based on a sample survey: sampling and nonsampling. The accuracy of a survey result
depends on both types of errors, but the full extent of the nonsampling error is unknown.
Consequently, particular care should be exercised in the interpretation of figures based on a
relatively small number of cases or on small differences between estimates. The standard errors
provided for the CPS/HVS estimates primarily indicate the magnitude of the sampling error.
They also partially measure the effect of some nonsampling errors in responses and enumeration;
but do not measure any systematic biases in the data. (Bias is the difference averaged over all
possible samples, between the estimate and the desired value.)
Nonsampling Variability
Nonsampling errors can be attributed to many sources, e.g., inability to obtain information
about all cases in the sample, definitional difficulties, differences in the interpretation of
questions, inability or unwillingness on the part of respondents to provide correct information,
inability to recall information, errors made in collection such as recording or coding the data,
errors made in processing the data, errors made in estimating values for missing data, and failure
to represent all units with the sample (undercoverage). Undercoverage in the CPS/HVS results
from missed housing units and misclassifying housing units. Ratio estimation to independent
controls, as described previously, partially corrects for the bias due to survey undercoverage.
However, biases exist in the estimates to the extent that missed households have different
characteristics than interviewed households.
Sampling Variability
The standard errors given in the following tables are primarily measures of sampling
variability, that is, of the variations that occurred by chance because a sample rather than the
entire population was surveyed. The sample estimate and its standard error enable one to
construct confidence intervals; ranges that would include the average results of all possible
samples with a known probability. For example, if all possible samples were selected, each of
these being surveyed under essentially the same general conditions and using the same sample
design, and if an estimate and its standard error were calculated from each sample, then
approximately 90 percent of the intervals from 1.6 standard errors below the estimate to 1.6
standard errors above the estimate would include the average result of all possible samples.
The average estimate derived from all possible samples is or is not contained in any
particular computed interval. However, for a particular sample, one can say with specified
confidence that the average estimate derived from all possible samples is included in the
confidence interval.
Standard errors may also be used to perform hypothesis testing, a procedure for
distinguishing between population parameters using sample estimates. The most common types
of hypotheses appearing in this report are: (1) the population parameters are identical, and, (2) the
population parameters are different. An example of this would be comparing the vacancy rate in
MA's versus the vacancy rate outside MA's. Tests may be performed at various levels of
significance, where a level of significance is the probability of concluding that the characteristics
are different when, in fact, they are identical.
To perform the most common test, let x and y be sample estimates for two characteristics
of interest. Let the standard error on the difference x-y be SEDIFF. If the ratio R = (x-y)/SEDIFF
is between -1.6 and +1.6, no conclusion about the difference between the characteristics is
justified at the 0.10 level of significance. If, on the other hand, this ratio is smaller than -1.6 or
larger than +1.6, the observed difference is significant at the 0.10 level. In this event, it is a
commonly accepted practice to say that the characteristics are different. Of course, sometimes this
conclusion will be wrong. When the characteristics are, in fact, the same, there is a 10 percent
chance of concluding that they are different. All statements of comparison in the text have passed
a hypothesis test at the 0.10 level of significance or better. This means that, for most differences
cited in the text, the estimated difference between characteristics is greater than 1.6 times the
standard error of the difference.
Comparisons of characteristics of vacancies for 1990 (which include year-round vacant
mobile homes as part of the year-round vacant inventory for the first time) with previous
unrevised years reveal significant differences in some cases. Thus caution should be used when
comparing current data with previous unrevised data prior to 1990.
Illustration of the Use of Tables of Standard Errors
The sample estimate and its standard error enable one to construct a confidence interval. A
confidence interval is a measure of an estimate's variability. The larger a confidence interval is in
relation to the size of the estimate, the less reliable the estimate. For example, the estimated
percent of all housing units vacant and available for rent in the Midwest is 3.2 percent (Table 7)
and the standard error on the estimate is 0.1 percentage points. Then the 90-percent confidence
interval (Table B-1) is calculated as 3.2 percent +/- (1.645 x 0.1), or 3.2 percent +/- 0.2, or from
3.0 percent to 3.4 percent. If all possible samples were surveyed under essentially the same
general conditions and the same sample design, and if an estimate and its standard error were
calculated from each sample, then approximately 90 percent of the intervals from 1.645 standard
errors below the estimate to1.645 standard errors above the estimate would include the average
result of all possible samples.
Standard errors are also used to perform hypothesis testing--a procedure for distinguishing
between population parameters (whose values are not known) using sample estimates. The most
common type of hypothesis is that two population parameters are different. These tests may be
performed at various significance levels. The significance level is the probability of concluding
that the parameters are different when, in fact, they are the same. For example, to conclude that
two parameters are different at the 0.10 level of significance, the absolute value of the difference
between their corresponding sample estimates must be greater than or equal to1.645 times the
estimated standard error of the difference.
The Census Bureau uses 90-percent confidence intervals and 0.10 levels of significance to
determine statistical validity. The 90-percent confidence intervals are shown in the text or in the
tables for selected items. The standard errors for other figures in this report are given in the tables.
In addition to sampling error, the figures in this report, both the estimates and their standard
errors, are also subject to rounding error.
Go to Housing Vacancies and Homeownership Annual Statistics: 2007
Contact Bob Callis or Linda Cavanaugh at (301)763-3199 or visit ask.census.gov for further information on the Housing Vacancy Survey.
Source: U.S. Census Bureau, Housing and Household Economic Statistics Division
Last Revised: February 20, 2008
