|
Chapter 1.
Labor Force Data Derived from the Current Population
Survey
Sampling
Since the inception of the survey, there have been various
changes in the design of the CPS sample. The sample is
traditionally redesigned and a new sample is selected after each
decennial census. Also, the number of sample areas and the
number of sample persons are changed occasionally. Most of these
changes are made in order to improve the efficiency of the sample
design, increase the reliability of the sample estimates, or
control costs. Since the mid-1980s, the CPS has had a State-based
sample design, meaning that all sampling operations such as
allocation and selection are implemented at the State level.
A redesigned CPS sample based on the 1990 decennial census was
selected for use during the 1990s and the early years of the new
century. Households from this new sample were phased into the
CPS between April 1994 and July 1995. The July 1995 sample was
the first monthly sample based entirely on the 1990 census. A
redesigned sample based on the results of the 2000 census will be
phased in between April 2004 and July 2005.
The original 1990 census-based sample design included about
66,000 housing units per month located in 792 selected geographic
areas called primary sampling units (PSUs). The sample initially
was selected to meet specific reliability criteria for the
Nation, for each of the 50 States and the District of Columbia,
and for the substate areas of New York City and the Los
Angeles-Long Beach metropolitan area. In 1996, the original
reliability criteria for the sample design were modified to
reduce costs, which decreased the sample to 754 PSUs and 59,000
housing units. The current criteria, given below, are based on
the coefficient of variation (CV) of the unemployment level,
where the CV is defined as the standard error of the estimate
divided by the estimate, expressed as a percentage. These CV
controls assume a 6-percent unemployment rate in order to
establish a consistent specification of sampling error.
The current sample design, including an expansion to meet the
requirements of the SCHIP legislation, was introduced in July
2001. It includes about 72,000 households from 754 sample areas,
or PSUs, and maintains a 1.9-percent CV on national monthly
estimates of unemployment level. This translates into a change of
0.2 percentage point in the unemployment rate being significant
at a 90-percent confidence level. For each of the 50 States and
for the District of Columbia, the design maintains a CV of at
most 8 percent on the annual average estimate of unemployment
level, assuming a 6-percent unemployment rate. Due to the
national reliability criterion, estimates for several large
States are substantially more reliable than the State design
criterion requires. Annual average unemployment estimates for
California, Florida, New York, and Texas, for example, carry a
CV of less than 4 percent.
In the first stage of sampling, the 754 PSUs are chosen. In
the second stage, ultimate sampling unit clusters composed of
about four housing units each are selected. Each month, about
72,000 housing units are assigned for data collection, of which
about 60,000 are occupied and thus eligible for interview. The
remainder are units found to be destroyed, vacant, converted to
nonresidential use, containing persons whose usual place of
residence is elsewhere, or ineligible for other reasons. Of the
60,000 housing units, about 7 to 8 percent are not interviewed
in a given month due to temporary absence (vacation, for example)
of the occupants, other failures to make contact after repeated
attempts, inability of persons contacted to respond,
unavailability for other reasons, and refusals to cooperate
(about half of the noninterviews). Information is obtained
each month for about 110,000 persons 16 years of age or older.
Selection of sample areas. The entire area of the United States, consisting of 3,141 counties and independent cities, is divided into 2,007 PSUs. PSUs are defined within States and do not cross State boundaries. In most States, a PSU consists of a county or a number of contiguous counties. In New England and Hawaii, minor civil divisions are used instead of counties.
Metropolitan areas within a State are used as a basis for
forming many PSUs. Outside of metropolitan areas, two or more
counties normally are combined to form a PSU except when the
geographic area of an individual county is too large. Combining
counties to form PSUs provides greater heterogeneity; a typical
PSU includes urban and rural residents of both high and low
economic levels and encompasses, to the extent feasible, diverse
occupations and industries. Another important consideration is
that the PSU be sufficiently compact so that, with a small sample
spread throughout, it can be efficiently canvassed without
undue travel costs.
The 2,007 PSUs are grouped into strata within each State.
Then, one PSU is selected from each stratum with the probability
of selection proportional to the population of the PSU.
Nationally, there are a total of 428 PSUs in strata by
themselves. These strata are self-representing and generally
are the most populous PSUs in each State. The 326 remaining
strata are formed by combining PSUs that are similar in such
characteristics as unemployment, proportion of housing units
with three or more persons, number of persons employed in
various industries, and average monthly wages for various
industries. The single PSU randomly selected from each of these
strata is non-self-representing because it represents not
only itself but the entire stratum. The probability of selecting
a particular PSU in a non-self-representing stratum is
proportional to its 1990 population. For example, within a
stratum, the chance that a PSU with a population of 50,000 would
be selected for the sample is twice that for a PSU having a
population of 25,000.
Selection of sample households. Because the sample design is
State based, the sampling ratio differs by State and depends on
State population size as well as both national and State
reliability requirements. The State sampling ratios range
roughly from 1 in every 200 households to 1 in every 3,000
households. The sampling ratio occasionally is modified slightly
to hold the size of the sample relatively constant given the
overall growth of the population (called “sample maintenance
reduction”). The sampling ratio used within a sample PSU
depends on the probability of selection of the PSU and the
sampling ratio for the State. In a sample PSU with a probability
of selection of 1 in 10 and a State sampling ratio of 3,000, a
within-PSU sampling ratio of 1 in 300 achieves the desired
overall ratio of 1 in 3,000 for the stratum.
The 1990 within-PSU sample design was developed using
block-level data from the 1990 census. (The 1990 census was the
first decennial census that produced data at the block level for
the entire country.) Normally, census blocks are bounded by
streets and other prominent physical features such as rivers or
railroad tracks. County, Minor Civil Division, and census place
limits also serve as block boundaries. In cities, blocks can be
bounded by four streets and be quite small in land area. In rural
areas, blocks can be several square miles in size.
For the purpose of sample selection, census blocks were
grouped into three strata: Unit, group quarters, and area.
(Occasionally, units within a block were split between the unit
and group-quarters strata.) The unit stratum contained regular
housing units with addresses that were easy to locate
(for example, most single-family homes, townhouses, condominiums,
apartment units, and mobile homes). The group-quarters stratum
contained housing units in which residents shared common
facilities or received formal or authorized care or custody.
Unit and group-quarters blocks exist primarily in urban and
suburban areas. The area stratum contains blocks with addresses
that are more difficult to locate. Area blocks exist primarily in
rural areas.
To reduce the variability of the survey estimates and to
ensure that the within-PSU sample would reflect the demographic
and socioeconomic characteristics of the PSU, blocks within the
unit, group-quarters, and area strata were sorted using
geographic and block-level data from the census. Examples of
the census variables used for sorting include proportion of
minority renter-occupied housing units, proportion of housing
units with female householders, and proportion of
owner-occupied housing units. The specific sorting variables
used differed by type of PSU (urban or rural) and stratum.
Within each block, housing units were sorted geographically
and grouped into clusters of approximately four units.
A systematic sample of these clusters was then selected
independently from each stratum using the appropriate
within-PSU sampling ratio. The geographic clustering of the
sample units reduces field representative travel costs. Prior to
the interview, special listing procedures are used to locate the
particular sample addresses in the group-quarters and area
blocks.
Units in the three strata described above all existed at the
time of the 1990 decennial census. Through a series of additional
procedures, a sample of building permits is included in the CPS
to represent housing units built after the decennial census.
Adding these newly built units keeps the sample up-to-date and
representative of the population. It also helps to keep the
sample size stable: over the life of the sample, the addition
of newly built housing units compensates for the loss of “old”
units that may be abandoned, demolished, or converted to
nonresidential use. In normal circumstances, the number of
eligible households in the sample grows slowly. Sample
maintenance reduction procedures are periodically implemented to
hold the size of the sample relatively constant.
Rotation of sample. Part of the sample is changed each month.
Each monthly sample is divided into eight representative
subsamples or rotation groups. A given rotation group is
interviewed for a total of 8 months, divided into two equal
periods. The group is in the sample for 4 consecutive months,
leaves the sample during the following 8 months, and then returns
for another 4 consecutive months. In each monthly sample, 1 of
the 8 rotation groups is in the first month of enumeration,
another rotation group is in the second month, and so on.
(The rotation group in the fifth month of enumeration is
returning after an 8-month break.) Under this system, 75 percent
of the sample is common from month to month and 50 percent is
common from year to year for the same month. This procedure
provides a substantial amount of month-to-month and year-to-year
overlap in the sample, thus yielding better estimates of change
and reducing discontinuities in the series of data without
burdening sampled households with an unduly long period of
inquiry.
Next: Collection Methods
|
