97-137 GOV
Updated April 27, 1998
CRS Report for Congress
Received through the CRS Web
Census 2000: The Sampling Debate
Jennifer D. Williams
Analyst in American National Government
Government Division
Summary
Plans by the Bureau of the Census to incorporate data from two new sample surveys
into the 2000 decennial census count have had a mixed congressional reception. Three
sampling bills in the 105 Congress (H.R. 1220, H.R. 1178, and H.R. 776) have bee
th
n
referred to committee, without further action. Sampling has been debated chiefly in the
appropriations process (H.R. 1469, H.R. 1871 [P.L. 105-18; 111 Stat. 158], and H.R.
2267/S. 1022 [P.L. 105-119; 111 Stat. 2440]). The bureau now is a defendant in two
anti-sampling suits brought under P.L. 105-119. The law also established a Census
Monitoring Board and directed the bureau to prepare for a traditional headcount in 2000,
not just to continue with its sampling plans. This report will be updated to reflect further
legislative action. For a related CRS product, see:
Census 2000: Sampling as an
Appropriations Issue in the 105 Congress.
th
Background
In 2000, the Bureau of the Census, Department of Commerce, will conduct the 22nd
consecutive decennial enumeration of the U.S. population. The count is a constitutional
requirement for reapportioning the House of Representatives. Census data also are used
for within-state legislative redistricting, allocating certain federal funds to states and
localities, administering various federal programs, and performing many kinds of research.
Two problems that underlie ongoing congressional concern about the bureau’s
preparations for Census 2000 are coverage (whom the census counts accurately and whom
it misses or counts incorrectly) and costs. Of equal concern, however, are the bureau’s
plans to deal with these problems by incorporating the results of two new sample surveys
into the 2000 enumeration.
Although the 1990 census counted about 98.4% of the total population, this census
was the first since 1940 in which coverage did not improve over that in the previous
enumeration. Further, the recurrent problem of the “differential undercount” (in which the
census inadvertently omits higher proportions of minority groups than of the majority
population) persisted in 1990. Estimates indicate that the 1990 census missed 4.4% of
blacks, 5.0% of Hispanics, 2.3% of Asians and Pacific Islanders, and 4.5% of American
Congressional Research Service ˜
The Library of Congress
CRS-2
Indians, compared with 0.7% of non-Hispanic whites. The difference between the black
and nonblack estimated undercounts was the largest since 1940.1
Despite repeated legislative and judicial attempts to require statistical “adjustment”
(alteration based on sample survey data) of the 1980 and 1990 census figures, the bureau
never officially has made such an adjustment.2 The stated purpose of these efforts was to
improve census accuracy. While maintaining that adjustment would not have met this goal
unequivocally for the past two censuses, the bureau has continued working to develop
adequate methods for coverage improvement and cost control in Census 2000.
The 1990 census cost a record-high $2.6 billion. In constant 1990 dollars, the cost
of enumeration rose from about $11.00 per housing unit in 1970 to $20.00 in 1980 and
$25.00 in 1990. The 1970 through 1990 censuses used essentially the same approach:
mailed-out, mailed-back questionnaires, with multiple followup visits by enumerators to
households (the occupants of housing units, one household per unit) whose members did
not return their questionnaires. The percentage of questionnaires filled out and returned
from all occupied housing units dropped from 87% in 1970 to 83% in 1980 and 74% in
1990.
3 Declining public cooperation with enumeration has meant greater expenditures for
nonresponse followup, estimated at $490 million to $560 million for the 1990 census. The
bureau has projected that a 2000 census incorporating the results of two proposed sample
surveys might cost almost $4.0 billion, but a census without these surveys might cost an
additional $675 million to $800 million.4
Congress, through P.L. 102-135; 105 Stat. 635, directed the Secretary of Commerce
to contract with the National Academy of Sciences (NAS) to study alternative census
methods and their costs. One resulting NAS report noted that “sampling for nonresponse
followup could produce major cost savings in 2000.” The report also endorsed sampling
at a second point in the census process, stating that the “differential undercount cannot be
reduced to acceptable levels at acceptable costs without the use of integrated coverage
measurement and the statistical methods associated with it.”5 (“Integrated coverage
measurement” is the bureau’s term for its proposed correction of coverage errors in the
2000 census.)
1 U.S. General Accounting Office,
Decennial Census: 1990 Results Show Need for Fundamental
Reform, GAO report GAO/GGD-92-94 (Washington: June 9, 1992), pp. 19-21; U.S. Bureau of
the Census,
Report to Congress: The Plan for Census 2000 (Washington: July 1997, reissued
Aug. 1997), p. 4.
2 For a history of the adjustment controversy, see: U.S. Library of Congress, Congressional
Research Service,
Decennial Census Coverage: The Adjustment Issue, by Jennifer D. Williams,
CRS Report 94-89 GOV (Washington: July 23, 1996).
3 U.S. General Accounting Office,
Decennial Census: 1990 Results Show Need for Fundamental
Reform, pp. 23-24, 36.
4 National Research Council, Committee on National Statistics, Panel to Evaluate Alternative
Census Methods,
Counting People in the Information Age (Washington: National Academy Press,
1994), pp. 97-98; U.S. Bureau of the Census,
Report to Congress, pp. 37, 39.
National Research Council,
5
Counting People in the Information Age, p. 206.
CRS-3
On February 28, 1996, the bureau unveiled its Census 2000 strategy,6 which reaffirms
NAS’s recommendations for sampling. The bureau’s original sampling plans, with later
modifications, and the congressional response to these plans are discussed below.
The Sampling Plans
Nonresponse Followup. The bureau analyzed various options for nonresponse
followup. An option highlighted on February 28, 1996, and in later congressional
testimony, stated that the bureau would interview by telephone as many nonresponding
households as possible, then would resort to the most expensive type of followup:
personal visits to the remaining unenumerated housing units. These visits were to continue
until the bureau obtained responses for 90% of the housing unit addresses in a given
county. At 90%, the bureau would complete nonresponse followup for the county by
surveying a random sample of housing unit addresses for which questionnaire responses
were lacking. Enumerators would visit these addresses and obtain information that the
bureau would use to estimate data for the other nonresponding households.7
On September 12, 1996, the bureau announced that, before starting to sample, it
would attain a 90% response rate at the
census tract level, “to provide more equitable and
accurate representation of...neighborhoods.”
8 (A census tract is a statistical subdivision
of a county. The typical tract has 2,500 to 8,000 persons with fairly similar economic
characteristics.)
9 On March 11, 1997, the bureau announced a further modification. It will
not reach a 90% response rate in every census tract before sampling the remaining 10%.
Instead, it has chosen an option called “direct sampling,” in which random sampling will
10
begin immediately, or “directly,” after the mail-back/phone-in phase of the census ends.
11
At this point, the bureau will calculate the response rate for each tract and use the rate to
determine what fraction of nonresponding households to sample. For example, if a tract
has a 60% (.60) mail-back/phone-in response rate and a 40% (.40) nonresponse rate, the
bureau will sample three-out-of-four (75%, or .75) of the nonresponding households (.75
times .40) to obtain a 90% (.90) total response rate for the tract: .75 times .40 equals .30;
6 U.S. Bureau of the Census,
The Plan for Census 2000 (Washington: Feb. 1996, reissued April
1996).
7
Ibid., pp. III-34, III-36; U.S. Congress, House Committee on Government Reform and Oversight,
Census 2000: The Challenge of the Count, hearing, 104th Cong., 2 sess., June 6, 199
nd
6
(Washington: GPO, 1996), p. 27.
8 U.S. Bureau of the Census, “Sampling and Estimation Update,” distributed at the 2000 Census
Advisory Committee Meeting (Washington: Sept. 26, 1996).
9 U.S. Bureau of the Census,
1990 Census of Population and Housing, Guide, Part B, Glossary,
1990 CPH-R-1B (Washington: GPO, 1993), p. 10.
10 U.S. Bureau of the Census, “Commerce Department’s Census Bureau Releases Refinements to
Plans for Census 2000,” press release CB97-C.01 (Washington: March 11, 1997), p. 1; U.S.
Bureau of the Census,
Report to Congress, p. 28.
11 The bureau plans to mail two waves of census questionnaires; a mailed notice or reminder will
precede each wave. Questionnaires also will be available in multiple public locations, such as
shopping malls, community centers, and schools, for persons who did not receive or who misplaced
their questionnaires. Anyone who would rather answer the questionnaire by telephone than by mail
will be able to do so by calling a toll-free number. (U.S. Bureau of the Census,
Report to
Congress, pp. 21-22.)
CRS-4
.30 plus .60 equals .90. Alternatively, if the mail-back/phone-in response rate for a tract
is 80%, with a 20% nonresponse rate, the sample will comprise one-in-two (.50 times .20)
nonresponding households: .50 times .20 equals .10; .10 plus .80 equals .90. By this
means, the bureau still expects to have contact (through mailed-back/phoned-in responses
and enumerators’ interviews for nonresponse followup) with 90% of all households in
every census tract. Using data from the sample, the bureau will estimate information for
the nonresponding households not sampled.
Integrated Coverage Measurement. The bureau will conduct a second sample
survey, for integrated coverage measurement (ICM). The ICM survey will allow th
12
e
bureau to estimate the extent to which the initial Census 2000 results (mailed-
back/phoned-in responses and enumerators’ interviews) correctly included, missed, or
erroneously counted housing units and persons. By applying these estimates to the initial
census figures, the bureau expects to correct coverage errors.
“Dual system estimation” (DSE), which the bureau used to evaluate the quality of the
1980 and 1990 censuses, is the method it has chosen to produce estimates for ICM. The
DSE method involves comparison of two independent data sets for the same geographic
areas. For ICM, the geographic areas will be a sample of blocks (the smallest units of
census geography). One independent data set will be persons in the ICM sample; the other
will be persons included in the initial phases of the census. Comparing these groups will
enable the bureau to determine, for the sample blocks, the proportion of housing units and
persons included or missed in each group. The bureau will resolve discrepancies between
the initial census results and responses to the original ICM interviews, usually through
followup ICM interviews, to arrive at the correct information.
The ICM sample will consist of about 750,000 housing unit addresses in 25,000
blocks. To derive the sample, the bureau first will classify the nation’s seven million
blocks in groups called “strata,” then will select blocks at random from every stratum. The
strata will be based on the similar characteristics of blocks, as determined in the 1990
census, and will not cross state boundaries. One stratum in a state, for example, might
comprise every block in large central cities with a 1990 census population of at least 30%
black renters and at least 10% Hispanic renters. The bureau will design strata and
“poststrata” (discussed below) both to be representative of state populations and to
correct coverage errors for hard-to-count groups and areas. Examples of some groups
counted less effectively than others in the 1990 census are blacks and Hispanics, American
Indians on reservations, renters (particularly in rural areas), and children under age 18.
After completing the ICM interviews, the bureau will sort the initial 2000 census
population into poststrata, or groups of persons with similar chances of having been
counted in the initial census phases. The poststrata will classify persons according to state
geographic subdivision (such as rural or urban), housing unit owner or renter status, sex,
age, race, and Hispanic ethnicity. DSE “estimation factors,” applied to the initial census
results, will provide estimates of the total number of housing units and persons in each
poststratum. Summing across poststrata will produce estimates of state totals, and the
sum of state totals will yield national totals. The bureau will use the state-level estimation
factors by poststrata to obtain housing unit and population estimates for all blocks in the
United States.
12
Ibid., pp. 29-32.
CRS-5
The Congressional Response
Congress’s interest in better census coverage and its determination to control
spiraling census costs are matched by congressional reservations about sampling for
nonresponse followup and for ICM. These issues have carried over from the 104th
Congress to the 105th. Especially in the House, but in the Senate as well, sampling tends
to have Democratic support and Republican opposition. The press has attributed this
difference to the perception that sampling will benefit Democrats in House
reapportionment and within-state legislative redistricting after the 2000 census.13
On September 18, 1996, the bureau’s House oversight committee, Government
Reform and Oversight, adopted a nonbinding report that stated: “The Bureau should not
use sampling methods to complete or adjust the actual enumeration” in 2000. The report
14
questioned the objectivity of the sampling design, the constitutionality of sampling and its
possibly negative effect on public confidence in the census, the quality of small-area data
obtained partly from sampling, and the operational feasibility of completing two complex
sample surveys by the statutory deadline for producing official census numbers. In
dissenting opinions, 18 committee members asserted that the report presented no
counterproposal to correct persistent census coverage errors and control census costs.
The report, with dissenting views, set the terms for further debate.
Thus far in the 105th Congress, three sampling bills have been referred to the House
Government Reform and Oversight Committee, without further action. (1) H.R. 1220
(Petri), introduced March 21, 1997, would prohibit using sampling or other statistical
procedures to determine state population for House reapportionment. Mr. Petri
introduced H.R. 3589, with the same prohibition, in the 104th Congress. (2) H.R. 1178
(Maloney), introduced March 20, 1997, would amend Title 13,
U.S. Code, the census
statute, to clarify that sampling may be used to improve decennial census accuracy. (3)
H.R. 776 (Meek), introduced February 13, 1997, would require the bureau to contact, by
mail or in person, 90% of the households in a census tract before sampling the other 10%
for nonresponse followup. Mrs. Meek introduced H.R. 3558 in the 104th Congress. (As
already discussed, the bureau plans to arrive at this 90% response rate by direct sampling.)
The main forum for the sampling debate in the 105 Congress has been th
th
e
appropriations process, beginning with FY1997 supplemental appropriations legislation
15
for disaster relief. President Clinton vetoed the first such bill approved by Congress, H.R.
1469, in part because it contained language to ban the use of sampling in any census to
13 “Census Funds Dispute,”
Washington Post, July 11, 1997, sec. A, p. 3; John Mercurio,
“Appropriators Freeze Census Spending on Sampling,”
Roll Call, July 14, 1997, pp. 10, 32;
Juliana Gruenwald, “Statistics Stir the Passions of Parties’ Boosters,”
Congressional Quarterly
Weekly Report, June 21, 1997, pp. 1434-1436.
14 U.S. Congress, House Committee on Government Reform and Oversight,
Sampling and
Statistical Adjustment in the Decennial Census: Fundamental Flaws, 104th Cong., 2 sess., H.
nd
Rept. 104-821 (Washington: GPO, 1996), p. 17.
1 For more information about the appropriations controversy, see: U.S. Library of Congress
5
,
Congressional Research Service,
Census 2000: Sampling as an Appropriations Issue in the 105th
Congress, by Jennifer D. Williams, CRS Report 98-321GOV (Washington: March 31, 1998).
CRS-6
determine the reapportionment population.16 A second bill passed by Congress, H.R.
1871, stipulated only that, within 30 days of enactment, the Commerce Department give
Congress a detailed report on its proposed methods for conducting the 2000 census
17
,
including estimates of sampling errors. The President signed this legislation on June 12,
1997 (P.L. 105-18; 111 Stat. 158).
Congress then resumed the sampling debate in FY1998 appropriations bills for
Commerce, Justice, and State, the Judiciary, and Related Agencies, H.R. 2267 and S.
1022. As enacted (P.L. 105-119; 111 Stat. 2440), section 209 of this legislation provided
for expedited judicial review of the bureau’s sampling plans, to determine their legality and
constitutionality. The bureau currently is a defendant in two suits brought under this
section. Both suits seek to prevent the use of sampling in the census to derive th
18
e
reapportionment population. Section 209 further instructed the bureau to prepare for a
traditional headcount in 2000, not simply to continue with its sampling plans. This section
also stated that in Census 2000 and “any dress rehearsal or simulation” in preparation for
the census, “the number of persons enumerated without using statistical methods must be
publicly available for all levels of census geography....” The data subject to this provision
include the official state populations used for reapportionment and redistricting. The
provision counters the bureau’s proposed “one-number census,” that is, a single set of
official census reapportionment numbers (for release by the December 31, 2000, legal
deadline), which would incorporate sampling for nonresponse followup and would be
altered on the basis of ICM. Finally, Section 210 of the FY1998 appropriations legislation
established a Census Monitoring Board to oversee all aspects of the coming enumeration
and report periodically to Congress. The eight-member board (with two members
appointed by the Senate majority leader, two by the Speaker of the House, and four by the
President) will go out of existence on September 30, 2001.
The bureau’s present operating conditions include simultaneous preparation for two
types of census (one with, and one without, sampling); heightened scrutiny from the
Census Monitoring Board and a new House Government and Reform census
subcommittee, created on November 13, 1997; and the lack of a permanent director
19
,
after Martha Farnsworth Riche’s resignation, effective January 31, 1998. James Holmes,
head of the bureau’s Atlanta regional office, was named acting bureau director on January
30, 1998.20
16 U.S. Congress, Conference Committee, 1997,
Making Emergency Supplemental Appropriations
for Recovery from Natural Disasters, and for Overseas Peacekeeping Efforts, conference report
to accompany H.R. 1469, H.Rept. 105-119, 105th Cong., 1st sess. (Washington: GPO, 1997), pp.
66-67.
U.S. Bureau of the Census,
17
Report to Congress.
18
Glavin v.
Clinton, no. 98-207-A (D.VA filed Feb. 12, 1998);
U.S. House of Representatives v.
U.S. Department of Commerce, no. 1:98CV00456 (D.DC filed Feb. 20, 1998).
19
Congressional Record, daily edition, vol. 143, Nov. 13, 1997, pp. H10790-H10793.
20 U.S. Dept. of Commerce, “Commerce Secretary Daley Names Acting Census Bureau Director,”
press release issued Jan. 30, 1998, http://www.census.gov/Press-Release/com98-01.html, visited
Feb. 23, 1998.