1. business and industrial
2. business operations
3. business plans

SAMPLE EXPANSION PLANS FOR THE CURRENT POPULATION SURVEY
Alan R. Tupek, U.S Bureau of Labor Statistics
Preston J. Waite, Lawrence S. Cahoon, U.S. Bureau of the Census
Alan R. Tupek, 441 G Street N.W., Washington, D.C. 20212
I. Introduction
The
Bureau
Current
planning
Population
improvements
Survey
(CPS)
to
the
1990 Decennial Census.
using
and implementing the expanded sample.
the
of
computer
the
estimates.
in
collection
for
in
minorities,
1989).
recall bias.
The final section of
The CPS sample expansion will provide more
design
com-
the
paper
discusses
prehensive analysis of the components of unemploy-
sample design research now in progress.
2. Sample Expansion Options
labor
socioeconomic
force.
affecting
Analysis
distinct
industry
composition
of
groups
special
of
the
The proposed CPS for the 1990's calls
situations
States
or minority concentrations or geographic
a l l 50 States and the D i s t r i c t of Columbia.
ColLection of the data in one week, as is now done,
estimates.
Increasing
the
more
reliable
is not o p e r a t i o n a l l y possible
reliability of State
annual
estimates
of
force for rural areas and some metropolitan
doubling
of
the
the survey.
CPS
implementation
is
proposed
expected
to
of
the
be
accomplished
extending the collection period to
CPS
an
These proposals have been considered
impractical from the viewpoint of c o l l e c t i o n and
processing resources. Hence, other options are
sample size from 55,000 to 105,000 eligible houseThe
the
in some cases, more than a doubting of the cost of
Plans call for a nearly
expansion
with
increase in sample. Previous options f o r redesign
involved large scale expansion in work toad and,
the
area estimates.
holds.
the
sample expansion to provide more reliable data for
data will also improve
based
for
on
location will be possible with the
labor
of
other
considerations and provides an overview of
ment and
the
resulting
We also discuss issues of incon-
the use of an expanded sample for State estimates.
Butz
reliable estimates for States and allow more
The effects may occur due to the phase
the expanded sample and by data collection
sistency of State and national estimates caused by
increased longitudinal use of the CPS data, and an
expansion of the CPS sample size (Plewes and
of
two weeks following the reference week,
assisted interviews,
methods to reduce coverage error
We also
describe efforts to minimize the effects on survey
The improvements include a
new questionnaire, automation
the sample design efficiency despite the two
years between implementing the basic sample design
in conjunction
with the redesign of the CPS sample following
environment
mize
of Labor Statistics and the Bureau
of the Census are
sample
considered.
by
The
alternative
we
have
selected
retains the 1-week reference period now in use but
additional
data collection
spreads
collection
occurs in the week following the reference week of
period.
We also considered several options invol-
the 12th of the month.
ving pooling data from m u l t i p l e reference weeks to
week.
Currently,
most
integrated sample
State
of
the
The new sample would be an
designed
to
and national estimates.
produce
lected
during
week.
the
week
sample
entire sample.
begin
in
research
1996,
into
will
occur
in
1994.
The
time
for
The
week"
data.
for
producing
There
"monthly"
data
ALL WEEKS--Each week of the
con-
for expanding the CPS sample and the rea-
ence week.
cuss
natively, some type of moving
major design issues in implementing the
We describe efforts to
by
considered
spreading the
year
interviewing
Under one option, sets of four or five
consecutive
two-phase approach.
various ways to
options
would occur using the preceding week as the refer-
sons for selecting the two-phase approach. We disthe
are
The primary
sample across the month are provided below.
with collection of data over a two-week period.
sidered
idea of such proposals is to pool data across
accomplish this.
further
options
across
month date back at least to the e a r l y 1960's.
erence
estimation and to address concerns
This paper describes some of the
a 2-week
weeks to produce "monthly" data, rather than "ref-
of the Larger sample is planned to
permitting
over
They are discussed in the Gordon Committee report.
the
Implementation of the redesign for
the basic CPS sample
implementation
from
data
Proposals for spreading the CPS sample
the
col-
following the reference
State estimates would be derived
the
obtain monthly estimates.
reliable
National estimates
would be based on the portion of the
of
maxi-
weeks
represent
the months.
average
scheme is possible for monthly estimates.
72
Alter-
estimation
While
providing
the
best "monthly" data, the
Alternatively, the week following the
week
of
all weeks option would likely delay publication of
the 12th can be the second reference week.
monthly
ence periods are then wholly within the month, and
estimates
two
(in
some
months, three)
weeks beyond the current schedule.
problems
of
overlap
of
In
reference
options.
The
4-(or
5-)
provide
periods into
other months would occur more often
other
addition,
than
week
in
data from the middle of the month.
the alternatives, this proposal
the
away
reference
from
of
the
advantage,
tual
elimination of recall bias.
change
est
from
the
current
l-week reference
That change is likely to cause the great-
effect on seasonal factors.
publication
timing
and
In addition, the
comparability
with
payroll employment survey are affected.
sample
It
is
spread
provides
further
tual
than
on
the
week
a
week
the
likely
have
week
This
(or
of
would
concern is
which
12th,
the impact on the seasonal factors due to a
reference week change is tess than under
the
but
week
There would
the
all
weeks
of
not
delay
full
able way in that direction. It would
entail
the
schedule.
first
publication
relative
There would be some
reference
week
to
months
contained
from the previous month.
payroll
schedule.
the
some
of
reinterview,
training,
for
other
in
favor
the
the
publication
of
the
consistency,
on
a
reference
date
period.
The extent of
expansion
(2)
the
to
discard
to
of
maintain
consistency
the
the
design
CPS,
issues have emerged.
-
for
four
the
does
not
two-phase
significant design
These are:
coincide
with
and
that
expansion.
12th.
current
One is the week
The week prior to the 12th can be
publication
We would maintain
schedule,
-
the
With
Therefore, it is not
the
delayed
phase
sample there are risks of
but the reference
the
series
selecfor the
possible
the introduction of the two-phase sample.
classify
likely be more than the 3- week option,
than the all week option.
be the ability to maintain
schedule
but
same
issues
of phase in effects.
less
A major advantage would
the
the
the
publication
in
due to both the redesign and with
some
The impact on seasonal factors would
to
in of the expanded
discontinuities
week would fail partially in the previous month in
cases.
in
historical
make both designs maximally efficient.
the
of
with other series,
The timing of the sample redesign
tion
them
reference weeks within the month.
the second reference week.
the
In gen-
two-phase option. Under this option, there are two
of
A
for the current
two-phase option. The primary con-
In developing
surveys,
to
made
considered three options, but the negatives
cerns were: (I) the need
opera-
and other necessary funcclosest
are
depends on the option implemented.
We
tions.
TWO WEEKS--This option is the
survey
the 12th for national
associated with them led us
It also allows the use of
the fourth week of the month
employment
Inclusion of
week
3. Major Design Issues
weeks.
period
Another primary
Comparisons
and (3) publication timing.
three
reference
data from the CPS and data from the payr-
tional viewpoint, this option seems optimal, as it
for
The
collection beyond the present collection period.
one or two days
However, from an
the
would allow us to use the field and CATI staff
CPS
will
eral, a delay of one week occurs for each week
current
where
CPS
monthly estimates is required.
the month, it goes a consider-
delay
in
They
estimates would extend the collection
week
provide
with
beyond
delay
does
the
oll employment series.
data
and there would be a larger sample with
three
recall
has the same reference period, and the same
between
a reference week of the 12th.
Using
the
options involve collecting
Changing
publ ication
all
Comparability with the payroll sur-
vey is better than the all week option,
representation
above
result in inconsistencies.
By centering collection on the week of the
proposal,
concep-
some impact on the seasonal patterns
nomic data.
two
earlier than under the 4-(or 5-) week pro-
also be more sample PSUs than with
However, the
the 12th is standard for collecting eco-
posal.
can not be released at the same time.
The
to a two week reference period could
of
weeks)
weeks option.
month.
and cause a discontinuity in the data series.
of the 12th.
publication
whole
we discuss in the next section, is
data for more than one reference week.
THREE WEEKS--Distribute sample across the three
centered
as
change
All
desirable.
fewer sample PSUs and less work load
option would allow
Among
furthest
bias problem.
for field staff every week of the year.
weeks
the
be as disruptive to our data users as
the
Also, the
is
is
the concept of providing data that are
truly representative
period of this option creates the greatest concepperiod.
Refer-
We
under the general issue
If both the redesign and
expansion occur jointly, we risk only one
discontinuity in the data series.
for CPS estimates and estimates from the
- Interviewing for a portion of the sample
take
payroll employment survey.
73
wilt
place two weeks after the reference week
rather than the usual one week later.
for
Hence,
the
two-phase
design.
However,
after the
respondents may have different recall patterns
implementation of the two-phase design, it will be
depending on the time of the interview.
more
This
efficient
to
interview some cases in these
may result in biases in the data and differen-
eleven States during the second week.
tial bias by State.
nally
Sample designated for the first week of inter-
only one week and form the basis
viewing is used to produce national estimates.
interviewer
The combined first and second week
that work over a two
used
for State estimates.
sample
Therefore, it will
be possible to produce two national
for
is
with
the
current
not
estimates
official
be
are
discussed
consistent
greater
for
week
period
sion from the redesign.
the efficiency of the
However,
detail
the
implementation
of
haps, in 1996.
Since both do
same
time,
it
the
is
system
given
not
occur
at
A carefully controlled phase in
the
cal
sample
PSU's from the 1990
efficiency
housing
Census
and
For
it
is
not
cost
years
later
phase design.
some
of
from
reselect
We use a
PSUs
a
ever,
statisti-
of
new
PSUs
of
the
two-phase design in
normal
redesign.
ways
different
Many features increase
There
are,
how-
to exert more control of the phase in
than is p o s s i b l e with a normal redesign.
However,
and
introduction
the risk of affecting the data.
maximum
As with a
effective to select new PSUs in
1994, hire interviewers for those
two
The
of the design the two-phase sample for
1996 should also involve the same steps.
rede-
procedure to maximize overlap between succes-
1996 contains several features that are
the
units within those
frame.
the
and new interviewers needed for the new design.
the restratification of the primary sampling units
of
of
occurs after each decennial census to
sive designs to minimize the number
The redesign of the sample in 1994 will involve
selection
CPS
minimize effects on the data.
efficiency of both simultaneously.
PSUs,
split.
for cost and design reasons to implement the
signed
possible to maximize the
(PSUs), the reselection of sample
that
5. Phase In Effects
the
two-phase expansion, per-
not
expan-
would have been possible if it had not been neces-
The current plans are for the implementation of
for
will
designs will not be as efficient as
two phase later than the redesign.
tater
1996
two-phase
sary
and
Spreading
in
The designs will maximize
4. The Timing Issue
1994,
origi-
determining
loads and PSU sizes.
below.
the redesigned CPS in
The
be interviewed in
The problems described here are a direct result
monthly national
in
work
will
of the decision to separate the
estimate developed from the first week only.
The issues
sample
result in inefficient interviewer work loads.
the same time period. The State estimates
from the full sample will
redesigned
will
then
result
redesign,
the
number of new PSU w i l l
the PSUs for the two-
ienced
Doing this would result in deleting
the new PSUs and adding a d d i t i o n a l PSUs.
two-phase
expansion
The
in new PSUs and new interviewers.
with
the
major expansion
be greater than that exper-
regular redesign because of the
of
the
survey.
We expect
to
The h i r i n g and t r a i n i n g of i n t e r v i e w e r s f o r just a
reduce the number of new interviewers by designing
two
and s e l e c t i n g the samples f o r the
year
period would be very c o s t l y .
to keep a l l of the PSUs selected in
1996 design,
cies b u i l t
of
when
in
the
inefficiencies
whether they will occur.
They
During
We do have a choice
will
will occur, but not
example
can
occur
the
redesign
in
of the expansion, there
in the
occur
during
the
period.
tation of two phase design, or they can be
second week with the phase
spread
Some interviews
sample.
Current plans are to max-
interview
week.
There
will
current
one
in
of
the
two-phase
be concurrent changes in the
imize the efficiency of the redesign sample, which
estimation methodology.
Such changes do not
mally
sample redesign.
That results in some inefficien-
the
within
PSU sample.
required
of the two-phase design.
match
two
week interviewing period.
nor-
We w i l l
these changes in much the
as we c o n t r o l
same
the phase in of a redesigned
There is one s i g n i f i c a n t
in
advantage that we have
the phase in of the two-phase sample that does
uses
not e x i s t
In the eleven
redesign,
largest States, it is not necessary to add
a
sample.
that
This is a par-
ticular problem in that the two-phase design
the
manner
of
The sample size within a
State for the basic redesign will not
during
carefully control
cies after implementation of the two-phase design.
A similar situation exists in the selection
occur
week
w i l l be s h i f t e d to the
starts
in 1994.
For
during the 1994 to 1996 period the i n t e r -
views w i l l
either
phase
be s o m e s h i f t s
during the 1994 - 1996 period, after the implemenover the entire period.
basic
and the sample expansion j o i n t l y .
then there must be some i n e f f i c i e n -
into both designs.
the
I f we are
1994
in
the
regular
redesign.
soon as we introduce the sample
sample
74
During
the
we use the sample data in estimation as
into
the
field.
The only exception is dunTny work loads assigned to
lion.
new interviewers for training.
during the reference week,
expansion,
it
is
With the two phase
not necessary to introduce all
the sample into the estimation process
with
the
introduction
the
reliability
concurrent
expansion
of the estimates.
is not necessary to
into
the
incorporate
opportunity
the
to
field,
improves
improvement
control
permitting
the
a
The
phase in, train the
responses
compared to the
and
two-phase approach, we will assess the impact
is
given
two
the
proxy-reconciled
weeks
respondents
after
the
interviewed
study,
we
will
determine
potential changes to the questionnaire.
precise
The field test for delayed recall is
months
use random digit dialing with interviewing
with previous month's actual responses for
This work showed that recall
much
larger
than
for a one week delay.
1994 through August 1995.
control
However,
group
beginning
early
after the reference week and
be
The test will
during
for
interviewing
for
the end of the reference week.
highly
tations
of
We recognize limi-
and the possible effect on the test results.
force status, an extended recall technique was not
therefore,
in
reducing
mean squared error.
This
study shows the need to explore alternative
tech-
There w i l l
be two
niques
to
reduce recall bias.
parts to the delayed recall
project.
The
second phase will involve a field test.
the
questionnaire
to
of
always
It may contain modifica-
minimize
effects
of
delayed
error.
recall.
A
proposal
for
laboratory
provided
by
validation
amount
in
the
We
will
order
to
One
Such a technique can yield
that
deviate
very
hours
to
get.
and
is
to
of
errors
debriefings
The final decision to proceed with
of
the
be
based
on
events
week in the minds of
are
to
The
alternative now under consideration is imple-
mentation of the two-week option described earlier
section 2.
reference
weeks.
effect the CPS
reinstate
The
conceptional
estimates.
changing rapidly.
7. Multiple Estimates Issues. In
Subjects
the
current
when
the
The
the
cued with plausible event categories that may
only
This option entails a conceptional
large
help the respondent recall
rela-
to the other error, then the expanded sample
and context of the reference
respondents.
of
will not be implemented with a two-phase approach.
worked
interview involves asking
is
imple-
comparison
in
various
the
expanded sample using the two-
change from current practice due to the
the
of
If the error due to delayed recall is large
tive
the
little from information
detailed questions. The purpose
are
Hence, the study will
Cohany 1989).
self-validation
obtain
recall bias on key
obtained from employer records (Edwards, Levine, &
The
for
recall error to other sources of error in the CPS.
use a selfcompare
tele-
sources,
to obtain indirect measures of recall
phase approach will
information
of error produced in one week and two week
recall.
data
respondents.
procedure
work exists.
of
Direct measures
reinterviews
mentation
major Limitation is in verifying
effect
difficult
include
would
the
characteristics.
We expect
respondents
tions which
external
The objective of the field test
estimates
The
be essentially the same as
1994.
from
instance, the une~)loyment insurance rolls.
tion
January
We,
to supplement the RDD portion of
identified
the new CPS questionnaire proposed for implementain
plan
the sample with a sample of persons without
phones
first
phase will involve laboratory investigations.
a
the random digit dialing (ROD) design
correlated month-to month variables, such as labor
effective
a
the week
test group to start on the second Sunday following
Work by
Bushery and Woltman (1979) showed that for
May
scheduled
for
to
With
quantification
status by comparing current responses for previous
may be serious for some variables.
As a
interview is not feasible.
recall bias on estimates of labor force
recall bias with one month delay is likely
one
reference week.
of response error due to the one week delay in the
Work by Perkins and Staff (1969 and 1970) meas-
bias
by
This con~Darison helps determine the
small laboratory samples,
potential modifications to the
the same respondent.
Fur-
obtained
in the CPS interview are
self-validated,
result of the initial
Also,
questionnaire in order to minimize any impact.
the
responses
and proxy reports.
amount of error for
ured
atten~)ts
conducting a reconciliation interview for the self
Before a final decision is made on implementing
evaluate
repeated
reports for each respondent and
proxy
6. Delayed Recall
will
told
Ample
soliciting
information.
we
is
of
interviewers and reduce potential data effects.
the
subject
ther verification
greater
of an interpolated week on response error.
the
to retrieve the proper information is given.
the official estimates as soon as interview-
ing begins in
when asked for hours worked
time and encouragement to make
Therefore, it
the
example,
there was a holiday on Monday of that week.
of sample into the field.
This can be done because
For
different
change will
effect
will
be
labor force composition is
a
system
that
uses one week of interviews for the national esti-
informa-
75
mates and two weeks of interviews
estimates,
there
are
produces estimates
for
for
the
9.
State
multiple
opportunities to
various
characteristics.
sampling.
current
This
sample
is
for
during
the
production
national labor force estimates.
week,
Ling
plans are to designate a national
sample for interviewing
first
boundaries.
design
country
first
week
and
then
one
based
on
the
full
the
likely not
variance
very
gains
important
aggregate
An
additional
estimates.
problem
context
With
of
the
State
areas
strata
of
the
w i t h i n States
Within each stratum,
one PSU with p r o b a b i l i t y p r o p o r t i o n a t e
of
a
sample
to
State estimates w i l l
of
a
sys-
c l u s t e r s of
provide
reliability.
wilt
to
of revised estimates,
Many States
The eleven largest States r e q u i r e no
sample, since the required r e l i a b i l i t y
remain the same f o r these States.
The f u l l
change
reliable
r e q u i r e the a d d i t i o n
a d d i t i o n a l sample to achieve the required
additional
with revised estimates
use
remaining
into
The second stage of sampling is
require
sample
month-to-month
cross
of a State-based supplemental sample.
as a result of any delayed recall bias.
exists in the
The
grouped
not
m e t r o p o l i t a n areas are
The expansion of the CPS
level
full
largest
selection
monthly
national estimates. The bias in the national estimates would likely be larger for the
samp-
roughly four housing u n i t s w i t h i n each PSU.
size.
would be small and
for
select
tematic
sample would have a slightly
sample
are
to s i z e .
The estimates
lower variance due to the increased
However,
we
the
week tater revise the
estimates based on the full sample.
primary
Because the CPS is
PSUs do
based on economic v a r i a b l e s .
permits us to produce national
for
The
always in sample.
of
the official State estimates.
a
of
These PSUs are g e n e r a l l y RSAs
a State-based design,
Sample
from both weeks of interviewing form the basis
u n i t s (PSUs).
stages
c l u s t e r g e o g r a p h i c a l l y the
three counties in r u r a l areas.
During the second
additional interviewing will occur.
Such
we
in the urban areas and u s u a l l y contain from one to
week.
of the official
estimates based on the sample designated
First,
counties in the United States i n t o
These estimates need not all be consistent.
The
Overview of the Sample DesiQns
The CPS design involves two d i s t i n c t
two-phase design w i l t
the c u r r e n t CPS design.
be very
similar
The major d i f f e r e n c e s
between preliminary estimate for the current month
are the increased sample and the spread of the
i n t e r v i e w i n g over the two week p e r i o d .
The s t r a t -
and revised estimates for the previous month.
ification
month-to-month change estimates are the difference
the
extent
Thus,
do
not
are
plan
force
w i l t be
will
State-based.
still
The
design
be embedded in the two-phase sample.
I t may
or may not use a State-based s t r a t i f i c a t i o n .
The
of c l u s t e r s of four housing
mates.
the
State
PSU
w i t h i n PSU sample w i t [ also be a systematic sample
to issue revised national esti-
This decision means that
within
national
will
estimates,
The
be a systematic sample of clus-
ters of four housing u n i t s .
also
in the interest of protecting the
integrity of the official labor
we
sample
that the bias levels are different in
the two estimates, the change estimates
biased.
To
under
estimates
way
on
units.
Research
is
the means f o r c o o r d i n a t i n g the two
produced from the full sample will be inconsistent
designs
and
optimizing
with the official national estimate. From a
between
the
two
designs.
There are no plans to
4-8-4
rotation
system
istical
change the
perspective it is not clear that the most
efficient estimator of a total is the sum
estimates
stat-
for
each component.
of
10. S t r a t i f i c a t i o n
However, this may
[t
is
to
provide
the
two-phase
design
in
research
related
1996. These consider-
design
will
The first
the
time
estimates,
area
mates
implementation of the two-phase design.
This area
District
includes issues of stratification, PSU, selection,
Design 2.
the
two
weeks
the
sample
design
from
monthly
is
1994
produce
for
the
to
National
State
redesign
1996.
tabor
This
force
estimates f o r the
the
ing.
between
The
second
It will
76
is
a
national
week of i n t e r v i e w -
subset
of
the
two-
produce n a t i o n a l e s t i -
mates at the current levels
design.
39 States and the
design
the f i r s t
This design is a
phase design.
of interviewing for the two-phase
remaining
of Columbia.
design u t i l i z i n g
The second area includes work
on the optimum allocation of
first
use
eleven largest States, and annual State e s t i -
delay
between the implementation of the redesign and the
and sample overlap.
CPS
and PSU
Desi9n 1. The
for
to
the
be necessary to develop three separate
sample
implementation
ations fall into two broad areas.
includes
will
of
the
for
designs f o r the two Phase CPS system.
a
brief overview of some of the research on detailed
design considerations for
allocation
Selection Considerations
Other Design Considerations
The purpose of this section
sample
sample as a part of the sample expansion.
the
be a problem for some data users.
8.
the
of
reliability.
labor force changes across States.
It will not produce State level estimates.
Design
3.
The
third design is the total two-
phase design.
This design will
State level estimates.
capacity for
produce
The
the
It will also have the
producing
national
We
expect
retaining
best
inefficiency
in
selecting
using
those
design.
PSUs
Thus,
and
we
for
in
that
for
same
Plans are to evaluate the recall
then one of the alternative approaches,
in
redesigned
the
paper,
As
we
all
designs
The authors would like to acknowledge Paul Mul-
the redesigned sample from 1994 to 1996
after
tin
1996.
simultaneously
is
not
the system.
We are studying
methods
of
PSUs for all of the designs.
ods would use
controlled
selection
jointly
the two-phase design is implemented.
PSU
selection
methodologies
guaran-
Edwards, W. Sherman, Levine, Roger and
Hours Worked".
Other design
are also under
Proceedings
of
efficiency
considerations
Spreading the
interviews
across
best
U.S.
the
strategy
design
is
to
spread
the
California to each week.
CPS."
half
the
the
sample
week of interviewing.
in
April 2, 1971.
. . . . . .
from
The research on
the
U.S.
Bureau
Commerce,
Bureau
of the
from
J. Bushery to H. gottman,
on
the
from
R.
on
Aquilino
Labor
to
Force
Data."
W. M. Perkins,
."Methods Test Phase Ill: Second Report on
the
randum
from
C.
Jones
and
Memo-
R. Aquilino to W. M.
Perkins, January 29, 1970.
century.
into
_ . , _ = .
Survey
design
error
in
of
the
randum
take
survey design improvements and
crovide our users with the ability to
assess
from
June 24, 1969.
the CPS estimates.
to
"Methods Test Phase Ill: First Report on
of Nonself Response on Labor Force Status."
improvements
Reduction in sampling error is necessary
•
the
Accuracy of Retrospective Interviewing and Effects
the CPS over the years and plans
nonsampling
advantage
(1989):
of Nonself Response on Labor Force Status."
for further improvements in the 1990's continue to
reduce
of
P.
Survey into the
Accuracy of Retrospective Interviewing and Effects
to meet the needs of our data users into
21st
of
of Changes in Respondent
The expansion of the CPS sample is necessary to
introduced
Memorandum
Memorandum
12. Summary
the
William
Population
Proceedings
Department
We
The national reliability
data
Current
Accuracy of Retrospective interviewing and effects
optimum allocation is addressing these issues.
continue
Association,
Survey Research
on
."Methods Test Phase Ill: Third Report
Unfortunately
requirements would not be met with the
first
Section
April 13, 1979.
sample evenly
PSUs.
constraints wilt not permit this.
cannot assign, for example,
the
the
Census. "Use of a Two-Week Reference Period in the
The main impetus for the two-phase design is
across the two weeks in all
the
Cohany,
of the Census Annual Research Conference.
the expectation that it will help in the hiring of
qualified interviewers. From this perspective
the
21st Century".
will
two weeks requires fewer interviewers for the survey.
American Statistical
of
Plewes, Thomas J., Butz,
"Moving
the allocation of the sample to week of
interview.
the
Methods.
Sample to Week of Interview
number
of
Sharon (1989): "Cognitive Strategies for Reporting
11. Optimum Allocation of the
A
Chandler
typing this paper.
consideration.
determine
of the Bureau of Labor
References
tee that no PSUs from the redesign be dropped when
and
Copeland
Bureau of the Census for her skill and patience in
The meth-
and
Kennon
We would also like to thank Sue
possible.
Research is underway to maximize the efficiency of
selecting
and
Statistics for their contributions to this paper.
mentioned in the previous section, meeting
criterion
described
will be selected for implementing
the sample expansion.
introduced
If the prob-
lem is serious relative to other sources of error,
Acknowledgment
both
effect
before a final decision is made on imple-
in 1996. Additionally, we want efficient
for
the
characteristics using the full and
sample in 1994 be a subset of the two phase sample
and the two-phase sample
the
State estimates, and the
mentation of the two-phase approach.
not
the two-phase
the
the
problem
PSUs and sample
sample
prefer
this
However, there continue to be
recall
partial samples.
There is consider-
units for the redesigned CPS in 1994 and then
delayed
in
implementing
need to explain and control the multiple estimates
by
the sample selected for the redesign in
1994 for the two-phase design.
able
of
achieved
described
to
concerns with this method, especially
estimates,
that minimum costs and minimization
are
approach,
solution
larger sample size.
but will not be used for that purpose.
of potential data effects
two-phase
paper, promotes a
the
77
L.
E.
Williams
to
Memo-
W. M. Perkins,