[House Report 104-821]
[From the U.S. Government Publishing Office]
Union Calendar No. 449
104th Congress, 2nd Session - - - - - - House Report 104-821
SAMPLING AND STATISTICAL ADJUSTMENT IN THE DECENNIAL CENSUS:
FUNDAMENTAL FLAWS
__________
FOURTEENTH REPORT
by the
COMMITTEE ON GOVERNMENT
REFORM AND OVERSIGHT
together with
ADDITIONAL AND DISSENTING VIEWS
September 24, 1996.--Committed to the Committee of the Whole House on
the State of the Union and ordered to be printed
COMMITTEE ON GOVERNMENT REFORM AND OVERSIGHT
WILLIAM F. CLINGER, Jr., Pennsylvania, Chairman
BENJAMIN A. GILMAN, New York CARDISS COLLINS, Illinois
DAN BURTON, Indiana HENRY A. WAXMAN, California
J. DENNIS HASTERT, Illinois TOM LANTOS, California
CONSTANCE A. MORELLA, Maryland ROBERT E. WISE, Jr., West Virginia
CHRISTOPHER SHAYS, Connecticut MAJOR R. OWENS, New York
STEVEN SCHIFF, New Mexico EDOLPHUS TOWNS, New York
ILEANA ROS-LEHTINEN, Florida JOHN M. SPRATT, Jr., South Carolina
WILLIAM H. ZELIFF, Jr., New LOUISE McINTOSH SLAUGHTER, New York
Hampshire PAUL E. KANJORSKI, Pennsylvania
JOHN M. McHUGH, New York GARY A. CONDIT, California
STEPHEN HORN, California COLLIN C. PETERSON, Minnesota
JOHN L. MICA, Florida KAREN L. THURMAN, Florida
PETER BLUTE, Massachusetts CAROLYN B. MALONEY, New York
THOMAS M. DAVIS, Virginia THOMAS M. BARRETT, Wisconsin
DAVID M. McINTOSH, Indiana BARBARA-ROSE COLLINS, Michigan
RANDY TATE, Washington ELEANOR HOLMES NORTON, District of
DICK CHRYSLER, Michigan Columbia
GIL GUTKNECHT, Minnesota JAMES P. MORAN, Virginia
MARK E. SOUDER, Indiana GENE GREEN, Texas
WILLIAM J. MARTINI, New Jersey CARRIE P. MEEK, Florida
JOE SCARBOROUGH, Florida CHAKA FATTAH, Pennsylvania
JOHN B. SHADEGG, Arizona BILL BREWSTER, Oklahoma
MICHAEL PATRICK FLANAGAN, Illinois TIM HOLDEN, Pennsylvania
CHARLES F. BASS, New Hampshire ELIJAH CUMMINGS, Maryland
STEVEN C. LaTOURETTE, Ohio ------
MARSHALL ``MARK'' SANFORD, South BERNARD SANDERS, Vermont (Independent)
Carolina
ROBERT L. EHRLICH, Jr., Maryland
SCOTT L. KLUG, Wisconsin
James L. Clarke, Staff Director
Kevin M. Sabo, General Counsel
Judith McCoy, Chief Clerk
Bud Myers, Minority Staff Director
LETTER OF TRANSMITTAL
----------
House of Representatives,
Washington, DC, September 24, 1996.
Hon. Newt Gingrich,
Speaker of the House of Representatives,
Washington, DC.
Dear Mr. Speaker: By direction of the Committee on
Government Reform and Oversight, I submit herewith the
committee's fourteenth report to the 104th Congress.
William F. Clinger, Jr., Chairman.
C O N T E N T S
__________
Page
I. Summary..........................................................1
II. Background.......................................................3
III. Findings.........................................................4
IV. Recommendations.................................................16
VIEWS
Clarifying comments of Hon. William F. Clinger, Jr............... 19
Additional views of Hon. Steven Schiff........................... 20
Additional views of Hon. Ileana Ros-Lehtinen..................... 21
Dissenting views of Hon. Cardiss Collins, Hon. Henry A. Waxman,
Hon. Tom Lantos, Hon. Major R. Owens, Hon. Edolphus Towns, Hon.
John M. Spratt, Jr., Hon. Louise McIntosh Slaughter, Hon. Gary
A. Condit, Hon. Bernard Sanders, Hon. Karen L. Thurman, Hon.
Carolyn B. Maloney, Hon. Barbara-Rose Collins, Hon. Eleanor
Holmes Norton, Hon. James P. Moran, Hon. Gene Green, Hon.
Carrie P. Meek, Hon. Chaka Fattah, and Hon. Elijah E. Cummings. 23
Additional views of Hon. Thomas M. Barrett....................... 33
Additional dissenting views of Hon. Carrie P. Meek............... 35
Union Calendar No. 449
104th Congress Report
HOUSE OF REPRESENTATIVES
2nd Session 104-821
_______________________________________________________________________
SAMPLING AND STATISTICAL ADJUSTMENT IN THE DECENNIAL CENSUS:
FUNDAMENTAL FLAWS
_______
September 24, 1996.--Committed to the Committee of the Whole House on
the State of the Union and ordered to be printed
_______________________________________________________________________
Mr. Clinger, from the Committee on Government Reform and Oversight,
submitted the following
FOURTEENTH REPORT
together with
ADDITIONAL AND DISSENTING VIEWS
On September 18, 1996, the Committee on Government Reform
and Oversight approved and adopted a report entitled ``Sampling
and Statistical Adjustment in the Decennial Census: Fundamental
Flaws.'' The chairman was directed to transmit a copy to the
Speaker of the House.
I. Summary
The decennial census is mandated by the Constitution in
order to apportion the Congress. Census data are used by every
State for congressional and State redistricting. They are also
used to enforce the Voting Rights Act. Numerous Federal and
State programs, distributing billions of dollars each year, use
decennial census data, or the intercensal estimates derived
therefrom, for their implementation.
On February 28, 1996, the U.S. Department of Commerce and
the Bureau of the Census publicly announced plans for a
``reengineered 2000 Census.'' The plans call for the use of
statistical methods in two separate instances: (1) to sample
and estimate the final 10 percent of the population failing to
respond in the actual enumeration\1\ (`sampling'), and (2) to
use a separate sample of houses to estimate those persons
missed in the actual enumeration and the sample for nonresponse
and revise it accordingly (`adjustment').
---------------------------------------------------------------------------
\1\ ``Actual enumeration,'' for purposes of this report, means a
count based upon physical evidence using methods such mail-out forms,
administrative records, and enumerator visits.
---------------------------------------------------------------------------
Statistical techniques have been used by the Census Bureau
to assess the accuracy of census counts since 1950, but have
never been used to ``correct'' the original number for use in
apportioning Congress.\2\
---------------------------------------------------------------------------
\2\ In the 1970 Census, the Census Bureau did rely on sampling to a
limited extent when, at the 11th hour, it was discovered that a number
of occupied housing units had been erroneously listed as vacant. This
was not repeated in subsequent censuses.
---------------------------------------------------------------------------
After the Secretary of Commerce decided in July 1991 not to
make a statistical adjustment to the 1990 Census, over 50
lawsuits erupted, culminating in the 1995 case considered by
the Supreme Court, United States v. City of New York. The
Court's decision, handed down in March 1996, upheld the
Secretary's decision.\3\
---------------------------------------------------------------------------
\3\ Supreme Court of the United States, Syllabus, Wisconsin v. City
of New York, et. al., No. 94-1614. Argued January 10, 1996--Decided
March 20, 1996.
---------------------------------------------------------------------------
The committee finds that the problems that surrounded the
issue of statistical adjustment in the 1990 Census also plague
the plans for the 2000 Census. This is compounded by the plans
to incorporate sampling to complete the actual enumeration.
Findings
1. Sampling/statistical adjustment are inherently
problematic given the subjectivity in the various decisions
comprising the methodology.
2. The legal provisions that concern the use of sampling
for apportionment purposes, both in the Constitution and in
Federal law, are variously interpreted.
3. The inherent uncertainties of sampling/statistical
adjustment may undermine public confidence in the decennial
census and reduce public participation.
4. The Commerce Department and the Census Bureau have not
clearly distinguished between the two statistical methods
proposed for the 2000 Census: (1) sampling for nonresponse
follow-up, and (2) sampling for the Integrated Coverage
Measurement (ICM); nor have they clarified issues of accuracy
respective to the two sampling techniques.
5. The sampling method for nonresponse follow-up introduces
additional error into the process and may compromise the
accuracy of small-area data which are important for
congressional and State legislative redistricting.
6. The complexity of the two different sampling techniques
being planned for the 2000 Census adds a great deal of risk to
the operational feasibility of the Bureau's current approach.
Recommendations
1. Congress should work to clarify existing Federal
statutes with regard to the use of sampling to make statistical
adjustments to the census for apportionment purposes.
2. The Bureau should not use sampling methods to complete
or adjust the actual enumeration of the 2000 Census which is
constitutionally mandated for purposes of apportionment.
3. The Department of Commerce and the Bureau of the Census
should prioritize the constitutional mandate of the decennial
census--apportionment of the House of Representatives.
4. The Bureau should emphasize and strengthen its
cooperative relationships with State and local elected
officials, as well as members of local organizations, who are
vital in helping increase response rates to the decennial
census.
5. The Bureau should strengthen its plans for a thorough
quality check of the 2000 Census and maintain open access to
all processes for internal and external review and analysis.
II. Background
On October 25, 1995, Congressman William H. Zeliff, Jr.,
chairman of the Subcommittee on National Security,
International Affairs, and Criminal Justice, held an oversight
hearing to examine testimony from Census Bureau officials
regarding their plans for conducting the 2000 decennial census.
Witnesses included Dr. Martha Riche (Director, Bureau of the
Census, U.S. Department of Commerce), Francis DeGeorge
(Inspector General, U.S. Department of Commerce), and Nye
Stevens (Director of Federal Management and Workforce Issues,
U.S. General Accounting Office). At the hearing, the Bureau
announced a number of new initiatives, including the use of
statistical sampling to complete the actual enumeration.\4\
---------------------------------------------------------------------------
\4\ Martha F. Riche, testimony on behalf of the Bureau of the
Census, before the House Government Reform and Oversight Subcommittee
on National Security, International Affairs, and Criminal Justice,
October 25, 1995.
---------------------------------------------------------------------------
On February 28, 1996, Commerce Department and Census Bureau
officials publicly announced that sampling would be used in two
different instances to compile the 2000 decennial census. The
first use of sampling would be for nonresponse follow-up. After
counting 90 percent of the population of a county by an actual
enumeration, a sample would be selected from the remaining
nonresponding addresses, at a rate of 1 in 10. (Sampling will
not be used to complete the enumeration on American Indian
reservations, in Alaska Native villages, in the Virgin Islands,
or in the Pacific Island territories.) \5\ The results of this
sample would then be used to estimate the remaining 10 percent
of the population.
---------------------------------------------------------------------------
\5\ The Plan For Census 2000, Bureau of the Census, U.S. Department
of Commerce, April 5, 1996, p. III-34.
---------------------------------------------------------------------------
The second use of sampling would take place essentially
after the Bureau has the first census number (based on the 90
percent actual enumeration plus the 10 percent sample for
nonrespondents.) This second sampling procedure, called an
``Integrated Coverage Measurement,'' or ICM, would begin with a
survey of 750,000 households. The Bureau would then match the
responses in the ICM survey to the initial census results, and
``equitably determine[s], for states and racial/ethnic groups,
the number of people and housing units missed or counted more
than once.'' \6\ The Bureau would then integrate these
statistical adjustments to the initial results to produce a
``one-number'' census by the legal deadline of December 31,
2000.\7\
---------------------------------------------------------------------------
\6\ The Plan, p. III-41.
\7\ December 31 of every decennial census year is the legal
deadline for submission of the census numbers by the Secretary of
Commerce to the Congress.
---------------------------------------------------------------------------
On February 29, 1996, the Committee on Government Reform
and Oversight held a hearing to gather testimony from Members
of Congress and outside experts regarding the Bureau's new
methodology. Witnesses included Senator Herb Kohl (D-WI),
Congressman Thomas Sawyer (D-OH), Congressman Thomas Petri (R-
WI), Bruce Chapman (president, Discovery Institute, Seattle,
WA), Dr. Barbara Bailar (vice president, Survey Research,
National Opinion Research Center), Dr. Steve Murdock (director,
Department of Rural Sociology, Texas A&M University), Dr.
Kenneth Wachter (professor of statistics and demography,
University of California at Berkeley), Dr. Charles Schultze
(senior fellow, the Brookings Institution), and Dr. James
Trussell (director, Office of Population Research, Princeton
University).
On June 6, 1996, the Committee on Government Reform and
Oversight held another hearing to air questions and concerns
about statistical methods planned for the Census 2000. The
witnesses were Dr. Everett Ehrlich (Undersecretary of Commerce
for Economic Affairs, U.S. Department of Commerce), and Dr.
Martha Riche (Director, Bureau of the Census, U.S. Department
of Commerce). Congressman Thomas Petri delivered a brief
statement on his bill, H.R. 3589, to prohibit the use of
sampling in the 2000 Census. However, he did not receive
questioning by Members of the committee.
III. Findings
1. Sampling/statistical adjustment are inherently problematic given the
subjectivity in the various decisions required
The committee is seriously concerned about the subjective
nature of sampling as an estimation technique. The basic
decision involved in any sampling methodology is the choice of
the sample itself. The determinations of exactly who is in the
sample and exactly what characteristics belong to those
individuals are inherently subjective. The population and
characteristics derived from a sample drawn by one person could
look quite different from those drawn by another person.
For ICM, the sampling universe is divided into ``post
strata''. Post strata are demographic subgroups with certain
characteristics such as ``black male renters age 30-49.'' An
undercount rate is estimated for each post stratum, then
assumed to hold constant across relatively large geographical
areas. Failures in these assumptions of constancy, called
``heterogeneity,'' caused major problems in the 1990 attempt to
statistically adjust the census using a sample of the
population, called a post-enumeration survey (PES).\8\ Dr.
Kenneth Wachter, who was a member of the Special Advisory Panel
on Census Adjustment of the Secretary of Commerce from 1989 to
1991, reminded the committee in his February testimony that
heterogeneity impaired the results of the 1990 adjustment
sample, and he warned that it is still a problem which remains
unsolved as we approach the year 2000.\9\
---------------------------------------------------------------------------
\8\ The 1990 adjustment was to be based on a post enumeration
survey (PES) of 150,000 households. A PES is a matching study in which
an independent sample of households are interviewed at some point after
Census Day. The information gathered is compared to census
questionnaires from those same households to determine whether each
person was correctly counted, missed, or double-counted in the census.
\9\ Kenneth Wachter, testimony before the House Government Reform
and Oversight Committee, February 29, 1996.
---------------------------------------------------------------------------
There is also the question of the adjustment methodology
itself. For example, in its plans to implement the 2000 census
ICM sample survey, the Bureau has yet to decide between two
distinct estimation methodologies, ``CensusPlus'' and ``Dual
System Estimation'' (DSE). In a Census Bureau document
detailing the results and decisions of the 1995 Census Test, a
discussion of these two methodologies gets to the heart of the
problem. According to the memo:
The potential sources of error and the implications
of the quality of the raw data for the two methods are
examined. The effects sometimes are different for the
two methods because the assumptions underlying them are
different.\10\ (Emphasis added)
---------------------------------------------------------------------------
\10\ Memorandum No. 46, ``1995 Census Test Results,'' by E. Ann
Vacca, Mary Mulry, Ruth Ann Killion, Decennial Statistical Studies
Division, April 1, 1996, p. 25.
In fact, in a final comparison between the two choices for
---------------------------------------------------------------------------
implementing the ICM, the Bureau memorandum states:
As designed and implemented in the 1995 Census Test,
dual system estimation (DSE) increases the estimate
after nonresponse follow-up in the traditionally
undercounted groups, primarily Blacks and renters,
while CensusPlus does not. However, both CensusPlus and
DSE increase the post nonresponse follow-up estimates
for Hispanics. Only DSE increases the estimate for
Asians and Pacific Islanders.\11\
---------------------------------------------------------------------------
\11\ Id., p. 38.
That a choice of estimation methodologies, or a change in
assumptions, can so directly change the results of the survey,
is very disturbing to the committee. The inherent subjectivity
of sampling, and choices of estimation methodologies, whereby
outcome can be manipulated by a change in assumptions, bears
directly upon issues of quality, accuracy, and fairness.
In deciding against adjustment in 1990, Commerce Secretary
Mosbacher expressed concern that ``adjustment would open the
door to political tampering with the census.'' \12\ Secretary
Mosbacher noted the important distinction between actual
enumeration and statistical adjustment. The unsettling danger
of statistical adjustment ``is that the choice of the
adjustment method selected by Bureau officials can make a
difference in apportionment, and the political outcome of that
choice can be known in advance,'' while ``the outcome of the
enumeration process cannot be directly affected in such a
way.'' \13\
---------------------------------------------------------------------------
\12\ Decision of the Secretary of Commerce on Whether a Statistical
Adjustment of the 1990 Census of Population and Housing Should be Made
for Coverage Deficiencies Resulting in an Overcount or Undercount of
the Population, July 22, 1991, 56 Federal Register 33582 at 33583.
\13\ Id.
---------------------------------------------------------------------------
2. The legal provisions that concern the use of statistical adjustment
for apportionment purposes, both in the Constitution and in
Federal law, are variously interpreted
There remains an important legal issue of whether sampling/
adjustment for apportionment among the States is actually
permitted by the Constitution. Article I calls for an ``actual
Enumeration . . .'' and Section 2 of the 14th Amendment reads:
``Representatives shall be apportioned among the several States
according to their respective numbers, counting the whole
number of persons in each State . . .''. In addition 13 U.S.C.,
Section 195 reads: ``Except for the determination of population
for purposes of apportionment of Representatives in Congress
among the several States, the Secretary shall, if he considers
it feasible, authorize the use of the statistical method known
as ``sampling'' in carrying out the provisions of this title.''
Interpretations vary with regard to both the Constitution and
Title 13.
In ``The Plan For Census 2000,'' the claim is made with
regard to statistical methods that, ``[w]e are on solid
Constitutional grounds. Our proposal will withstand all legal
challenges . . .'' \14\ The committee is concerned about the
use of such claims when the issue of the constitutionality of
statistical methods has never been decided by the Supreme
Court--the final court for constitutional questions. In
addition, the issue of sampling for nonresponse to complete the
actual enumeration has never been before any court.
---------------------------------------------------------------------------
\14\ The Plan, p. II-6.
---------------------------------------------------------------------------
3. The inherent uncertainties of sampling/statistical adjustment may
undermine public confidence in the decennial census and reduce
public participation
Two concerns Secretary Mosbacher raised in his decision not
to adjust the 1990 Census were that: (1) the uncertainty of
sampling and the potential for political manipulation would
erode public confidence in the census numbers; and (2) by
making a statistical adjustment, participation would decline
both at the State and local levels and at the individual
level.\15\ Committee hearings have shown that these concerns
still remain.
---------------------------------------------------------------------------
\15\ Secretary Mosbacher's Decision, 56 Federal Register 33582.
---------------------------------------------------------------------------
Sampling/Adjustment May Erode Public Confidence
In deciding against adjustment, Secretary Mosbacher found
it ``unsettling that a subjective choice of statistical
methodology can create such a dramatic practical difference in
apportionment.'' \16\ The shortcomings generally inherent in
statistical formulations are that assumptions are the basis of
any statistical formula attempting to establish precise
populations. These assumptions are subjectively chosen and
weighed, and thus potentially wrong.\17\ In addition, adjusted
numbers are no more than estimates and--unlike the actual
enumeration, which is based on some verifiable physical
evidence--entirely the product of statistical inferences with
no physical evidence for verification.\18\
---------------------------------------------------------------------------
\16\ Id., at 33583.
\17\ Brief of U.S. Senators Herb Kohl, Arlen Specter and Russell
Feingold, as Amici Curiae in Support of Petitioners, United States
Department of Commerce, et. al., v. City of New York, et. al., November
9, 1995, p. 18.
\18\ Id.
---------------------------------------------------------------------------
Because the assumptions on which the statistical
adjustments rest are subjective, they are changeable. In
comparing the two post-1990 census adjustment results released
by the Bureau, it was found that minor technical differences
caused substantial differences in results. The differences in
terms of apportionment were extraordinary. Under one method,
two seats in the House of Representatives moved, while under
the other plan, only one seat moved.\19\
---------------------------------------------------------------------------
\19\ Secretary Mosbacher's Decision, 56 Federal Register 33582 at
33583.
---------------------------------------------------------------------------
According to one expert reviewing post-1990 adjustment
alternatives, among five reasonable alternative estimation
methods, none of the resulting apportionments of the House of
Representatives were the same. Eleven different States either
lost or gained a seat in at least one of the five models.\20\
---------------------------------------------------------------------------
\20\ Id.
---------------------------------------------------------------------------
In the committee's February 29, 1996 hearing, several
witnesses expressed concerns that the inherent uncertainties in
sampling/adjustment may erode public confidence in the census
numbers. Bruce Chapman, former Census Bureau Director under
President Reagan, addressed the need to ensure that public
perception is a key component of the Census 2000 plan.
The current Census Director herself has spoken of the
three legs of the census stool that must dictate the
process: cost, accuracy, and public perception. The
first two legs are important, as I have acknowledged,
but if the latter breaks, the whole construction comes
down. The term ``public perception'' could also be
described as trustworthiness. In a time when public
mistrust of government is rife, I question a change
that would introduce the invention of statistical
persons into the census--robots constructed of sampled
data and intellectual abstractions--to stand in the
place of real human beings.\21\
---------------------------------------------------------------------------
\21\ Bruce Chapman, testimony before the House Government Reform
and Oversight Committee, February 29, 1996.
Congressman Tom Sawyer (D-OH) also underscored the importance
---------------------------------------------------------------------------
of public confidence:
I think that the underlying question that touches
every one of the concerns that has been raised is one
of confidence. If the count of the nation does not
enjoy the confidence of the people that are being
counted, it will not work, no matter what techniques,
technologies, or other kinds of re-engineering take
place.\22\
---------------------------------------------------------------------------
\22\ Congressman Thomas Sawyer, testimony before the House
Government Reform and Oversight Committee, February 29, 1996.
---------------------------------------------------------------------------
A One-Number Census Will Not Avert Potential Criticism & Lawsuits
``For 1990, the release of the figures from the original
enumeration and from the Bureau's statistical procedures and
evaluations were available for independent analysis.'' \23\
Skepticism and criticism of the statistical procedures became
public knowledge from many of those independent reviews. In
addition, over 50 lawsuits were subsequently filed over the
1990 census adjustment. However, the Bureau believes that by
incorporating the sampling/adjustment procedures into the
actual enumeration to produce a ``one-number census,'' and
avoiding a two-number census, such comparisons will be
precluded, lawsuits will be avoided, and public confidence will
not be undermined.
---------------------------------------------------------------------------
\23\ Kenneth Wachter, testimony before the House Government Reform
and Oversight Committee, February 29, 1996.
---------------------------------------------------------------------------
In Bureau Director Martha Riche's October 25 testimony, she
stated, ``we must produce a `one-number census' that is right
the first time and allows the decennial results to be
determined by statisticians at the Census Bureau, not by
lawyers and judges.'' \24\
---------------------------------------------------------------------------
\24\ Martha Riche, testimony before House Committee on Government
Reform and Oversight Subcommittee on National Security, International
Affairs, and Criminal Justice, October 25, 1996.
---------------------------------------------------------------------------
Indeed, it is the thinking of many that after the 2000
Census, there will be, as there was after the 1990 census, a
flood of lawsuits because of the controversial use of sampling
to adjust the actual enumeration. Congressman Petri stated in
his oral testimony before the committee on February 29, 1996
that:
It [adjustment] will not settle or end litigation; it
will just add to litigation, because you will have a
floating undercount estimated number added to the basic
head count, which then will be allocated on a political
basis and lead to endless litigation and uncertainty,
meanwhile undermining the integrity of the whole
process, increasing cynicism, and reducing
participation in future censuses.\25\
---------------------------------------------------------------------------
\25\ Congressman Petri, testimony on February 29, 1996.
Former Census Bureau Director Bruce Chapman echoed the
belief that more litigation will result from the Bureau's
efforts to statistically adjust the 2000 Census. Referencing a
news article with the Bureau's claim that an adjusted ``one-
number'' census would help the Bureau get past the legal
problems, he warned that, ``. . . you at least ought to
consider the possibility that you will have far more lawsuits
in the future if you undertake sampling.'' \26\
---------------------------------------------------------------------------
\26\ Bruce Chapman, oral testimony on February 29, 1996.
---------------------------------------------------------------------------
Somewhat contradictory to its claim that lawyers and judges
will not be determining the census totals, the Bureau's Plan
for Census 2000 acknowledges that the Bureau must allocate
funding for anticipated lawsuits because, ``regardless of the
census design chosen, the Census Bureau must prepare for legal
challenges to census results and procedures.'' \27\
---------------------------------------------------------------------------
\27\ The Plan, p. III-55.
---------------------------------------------------------------------------
Sampling/Adjustment May Discourage State and Local Participation
In deciding not to adjust the 1990 census, Commerce
Secretary Mosbacher expressed concern that adjustment would
remove the incentive of States and localities to join in the
effort to get a full and complete count.\28\ Historically, the
Census Bureau has relied extensively on State and local leaders
to encourage census participation. Acknowledging that it
``cannot accomplish its goals for Census 2000 working
alone'',\29\ the Bureau states that it ``must build
partnerships with state, local, and tribal governments and
community groups.'' \30\
---------------------------------------------------------------------------
\28\ Secretary Mosbacher's Decision, 56 Federal Register 33582 at
33584.
\29\ The Plan, p. II-2.
\30\ Id.
---------------------------------------------------------------------------
Because the census numbers are the basis for political
representation at all levels and Federal funding allocations,
States and localities have a vital interest in achieving the
highest participation rates possible under traditional
enumeration methods.\31\ However, if civic leaders and public
officials believe that statistical adjustments will remedy
undercounts, it will be difficult for them to justify expending
large amounts of money and resources on promotion and outreach
programs. Without their support in creating public awareness
and a sense of involvement in the census, participation is
likely to decline further.
---------------------------------------------------------------------------
\31\ Secretary Mosbacher's Decision, 56 Federal Register 33582 at
33584.
---------------------------------------------------------------------------
Committee members and witnesses at the February 29, 1996
hearing also expressed concerns that sampling/adjustment would
discourage both State and local promotion and outreach efforts
as well as individual citizen participation. Congressman Thomas
Barrett (D-WI) stated in questioning the panel of Congressmen
that, ``the attitude that I think many Wisconsinites have on
the attempts for the post-census adjustment was that we got
these beautiful awards from the Federal Government telling us
what a great job we had done and how proud they were of us, and
that they were going to take away millions of dollars and [a]
congressional seat as a result of the fine job that we did.''
\32\
---------------------------------------------------------------------------
\32\ Congressman Thomas Barrett, in questioning witnesses at the
House Government Reform and Oversight Committee hearing, February 29,
1996. The State of Wisconsin conducted a statewide public awareness
campaign and a targeted outreach program which included a matching
grant program aimed at traditionally undercounted groups. As a result
of State and local efforts, Wisconsin had the highest voluntary mail
response rate in the country--75 percent] compared to 64 percent
nationwide. The Census Bureau formally recognized Wisconsin's efforts.
Despite the accomplishment, the post-census adjustment proposed by the
Census Bureau would have caused Wisconsin to lose a seat and a portion
of its Federal funds.
---------------------------------------------------------------------------
In response to Congressman Barrett's question of whether
the Bureau's move toward a sampling/adjustment approach creates
disincentives for States to conduct promotion and outreach
efforts, Congressman Thomas Petri (R-WI), asked rhetorically,
``if the adjustment is going to be done in any event, why
should they spend any money at the local level to encourage
compliance with the census procedures?'' \33\
---------------------------------------------------------------------------
\33\ Congressman Thomas Petri, testimony before the House
Government Reform and Oversight Committee, February 29, 1996.
---------------------------------------------------------------------------
Congressman Sawyer agreed with Congressmen Barrett and
Petri that ``that kind of response, that kind of participation
should always be rewarded in a democracy, and [he] would not
want to do anything to undermine the importance of that in the
public mind.'' \34\ Congressman Petri later stated that:
---------------------------------------------------------------------------
\34\ Sawyer, testimony, February 29, 1996.
We have an important national obligation, as a
Congress, to attempt to overcome [the declining
response rate] through the best public relations
campaign we can do of emphasizing to people that this
is a responsibility and privilege of citizenship to be
counted, and, if we think that people are not
participating in it, to reach out and communicate more,
and send people by, or try to get in touch with their
communities, and use 101 techniques to involve them in
the process, not just to give up and then try to adjust
it through the political process and think that we've
accomplished something.\35\
---------------------------------------------------------------------------
\35\ Petri, testimony, February 29, 1996.
---------------------------------------------------------------------------
Sampling/Adjustment May Cause Further Decline in Public Participation
Voluntary participation is the cornerstone of the decennial
census because voluntary public response through mail-back is
the most accurate, effective and efficient source of census
data.\36\ The significance of a large mail-back response it not
just that it reduces the staff, time and money required, but
that it produces the best quality census data.\37\ When
voluntary participation in the census declines, the costs rise
exponentially while accuracy decreases. Each percentage point
of mail-out nonresponse will cost an additional $25 million for
the 2000 Census which is an $8 million increase over the cost
of the 1990 Census nonresponse rate. Accordingly, a ``high
level of public cooperation is the key to obtaining accurate
data at a reasonable cost.'' \38\
---------------------------------------------------------------------------
\36\ General Accounting Office, ``Decennial Census: 1990 Results
Show Need for Fundamental Reform,'' GAO/GGD-92-94, June 1992, p. 35.
\37\ Id., p. 36.
\38\ Id., p. 35.
---------------------------------------------------------------------------
Although sampling for nonresponse to complete the count may
reduce the need for staff and reduce costs, the use of
statistical adjustments may discourage citizen participation
and erode public confidence in the census numbers. While the
Bureau's research shows that a number of demographic,
socioeconomic and attitudinal variables affect public response
rates, Bureau data also suggests that the public's willingness
to cooperate is, at least in part, within the Bureau's
control.\39\ However, sampling/adjustment is not an effort
directed at increasing response rates, it simply is a response
to declining public participation. The census questionnaire's
length, complexity and intrusiveness have all been identified
as discouraging respondents from completing the form. The
committee is concerned that, if individuals believe they will
be counted regardless, they will not expend the time and effort
to respond, causing what Undersecretary Ehrlich referred to as
``participation meltdown.'' \40\
---------------------------------------------------------------------------
\39\ Id., p. 5. The Bureau is planning a number of improvements to
increase response rates, i.e., more user-friendly forms, multiple
mailings, reminder cards, making the forms more available, and forming
partnerships with State, local, and community groups.
\40\ Everett M. Ehrlich, oral testimony before the Committee on
Government Reform and Oversight, June 6, 1996.
---------------------------------------------------------------------------
4. The Department of Commerce and the Census Bureau have not clearly
distinguished between the two statistical methods proposed for
the 2000 Census: (1) sampling for nonresponse follow-up, and
(2) sampling for the Integrated Coverage Measurement (ICM); nor
have they clarified issues of accuracy respective to the each
of these sampling techniques
In the public document entitled ``Plan for Census 2000,''
revised April 5, 1996, the Department and Bureau claim that
sampling for nonresponse follow-up will reduce cost and improve
accuracy. Specifically, on page II-5, the public is guaranteed
that this sampling technique will ``ensure that Census 2000 is
built around a solid core of field results, while reducing the
cost and improving the accuracy of the data on the final
increment of the population.'' (Emphasis added) This statement
would have the public believe that the 10 percent sample of the
hardest to reach populations will increase the accuracy of the
count for these populations over a 100-percent physical
enumeration.
However, on May 17, 1996, at a meeting of the Census 2000
Advisory Committee, Bureau Director Martha Riche stated that
the 10 percent sampling for nonrespondents was only meant to
address cost concerns; it was not an effort to address the
differential undercount.\41\ This statement was reiterated by
other Bureau officials on May 24, 1996, in a meeting with
committee majority and minority staff. At that meeting,
minority staff Ben Cohen questioned Bureau experts about
whether their sample for nonresponse would improve the
differential undercount. The Bureau's response was that they
had no hard evidence that quality would be improved, but they
``assume'' the quality would be improved; that the sample for
nonresponse was primarily an effort to save money.\42\
---------------------------------------------------------------------------
\41\ House Government Reform and Oversight Committee staff notes,
Meeting of the Census 2000 Advisory Committee, May 17, 1996.
\42\ The meeting on May 24 included committee majority staff, Jane
Cobb; minority staff, David McMillen, staff of Representative Carrie
Meek, Ben Cohen; and Census Bureau officials Robert Marx, Paula
Schneider, and John Thompson, among others.
---------------------------------------------------------------------------
Congresswoman Carrie Meek (D-FL), at the February hearing,
stated that it concerned her that cost seemed to be the driving
factor behind the push for sampling for nonresponse follow-up:
``I am very concerned that one of the motives here, Mr. Sawyer,
may be to save money. I don't think this is the place we can
save money . . .''.\43\ Indeed, as Bureau officials have
stated, saving money has been a primary element driving the
Bureau toward use of sampling for nonresponse follow-up.
---------------------------------------------------------------------------
\43\ Congresswoman Carrie Meek, oral comments, House Government
Reform and Oversight Committee hearing, February 29, 1996.
---------------------------------------------------------------------------
5. The sampling method for nonresponse follow-up introduces additional
error into the process and may compromise the accuracy of
small-area data which is important for congressional and State
legislative redistricting
In hearings before the committee, concerns were raised by
both Members and outside experts regarding the introduction of
sampling error into the count. At the October 25, 1995 hearing
of the Subcommittee on National Security, International
Affairs, and Criminal Justice, GAO testified that, ``[t]he
nature of sampling itself, however, increases the statistical
uncertainty of the data on nonrespondents at lower geographic
levels. The magnitude of statistical uncertainty is dependent
on the size of the sample, the method used to draw the sample,
and the size of the universe being sampled.'' \44\ In other
words, the uncertainty of the sampling methodology for
nonresponse follow-up is inherent given the indeterminate
configuration of the nonresponding households which comprise
that particular sample universe and the method used to select
the sample households.
---------------------------------------------------------------------------
\44\ L. Nye Stevens, testimony of the General Accounting Office
before the House Government Reform and Oversight Committee,
Subcommittee on National Security, International Affairs, and Criminal
Justice, October 25, 1995.
---------------------------------------------------------------------------
The Bureau itself admits the negative implications of the
10-percent sample for nonresponse follow-up. ``The Plan For
Census 2000'' indicates:
Visiting only a sample of nonresponse housing units
raises issues of equity, reliability, and the size of
the margin of uncertainty associated with the totals
obtained. These issues are particularly important for
small population groups and for populations in small
towns and neighborhoods.\45\
---------------------------------------------------------------------------
\45\ The Plan, p. III-35.
The problem of accuracy was identified in testimony
received at the October 25, 1995 hearing of the Subcommittee on
National Security, International Affairs, and Criminal Justice.
In his evaluation of the Bureau's plans to conduct the 2000
Census, Commerce Inspector General Francis DeGeorge testified
that, ``using a 1-percent sample of the entire population to
represent the last 10 percent may introduce statistical
uncertainty, producing lower quality information. At 90
percent, the people who have not yet responded are the hardest
to count, so a disproportionate number of them will be
represented in the sample, possibly leading to missing
information and thus introducing statistical bias.'' \46\
---------------------------------------------------------------------------
\46\ Francis D. DeGeorge, testimony of the Inspector General of the
Department of Commerce before the House Government Reform and Oversight
Committee, Subcommittee on National Security, International Affairs,
and Criminal Justice, October 25, 1995.
---------------------------------------------------------------------------
In its April 5, 1996 version of ``The Plan For Census
2000,'' the Bureau admits also that, ``[s]ampling for
nonresponse will increase the margin of uncertainty for
estimates obtained from the sample form data items.'' \47\ The
sample form, or ``long form,'' which gathers socio-economic
information, goes to one in six households during the decennial
census.
---------------------------------------------------------------------------
\47\ The Plan, p. III-35.
---------------------------------------------------------------------------
In testimony received by the committee on February 29,
1996, issues of accuracy and the erosion of data quality were
major concerns among the witnesses. Dr. Robert Murdock, a rural
demographer and professor at Texas A&M University, testified
that:
I am particularly concerned about this issue from our
experience base in working in rural areas across the
south, because the areas that we have found it most
difficult to get an adequate sampling frame have been
those areas that have hard to enumerate populations,
particularly rural minority populations. In sum, about
this issue, I am concerned that the use of sampling to
complete the census count will lead to a degradation in
the quality of statistics, the quality of data from the
census for small rural areas. According to Census' own
figures, 48 percent of the 39,000 governmental units in
the United States in 1990 had populations of less than
1,000. And two-thirds, more than 67 percent, had less
than 2,500 persons.\48\
---------------------------------------------------------------------------
\48\ Robert Murdock, oral testimony before the House Government
Reform and Oversight Committee, February 29, 1996.
Dr. Murdock observed further in his written statement that,
``[s]ince these elements of inaccuracy would, in turn, likely
disproportionately impact minority racial/ethnic groups, such
problems could also impact the capability of the Integrated
Coverage Measurement procedures to produce accurate estimates
of the level of undercount.'' \49\
---------------------------------------------------------------------------
\49\ Murdock, testimony, February 29, 1996.
---------------------------------------------------------------------------
At the same hearing, Congressman Sawyer also expressed
concerns about losing quality data with sampling. He states in
his written testimony that:
I am also concerned about the Bureau's statement that
it plans to hold long form distribution to `smallest
possible sample of households'. . . . If, however, the
sample is as small as it can be, I have increased
concerns about the use of statistical methods in the
completion of nonresponse follow-up in the final 10
percent of households. Long-form data is used for a
wide array of applications. It forms the basis of
important economic and societal models. The private-
sector uses the data for broadly public goods--where to
place a 500-job factory, for instance. And local
governments--the largest non-Federal user of census--
use that information to plan traffic patterns, locate
schools and do critical urban planning. A suspect
result will compromise expert and public confidence in
these numbers that the nation needs to make myriad
decisions on long-term planning. . . . Clearly, there
are trade-offs in the use of statistical methods. The
Bureau must work to limit any potential loss in the
quality of the data.\50\
---------------------------------------------------------------------------
\50\ Thomas Sawyer, testimony before the Committee on Government
Reform and Oversight, February 29, 1996.
Dr. Charles Schultze, chair of the National Research
Council's ``Panel on Census Requirements in the Year 2000 and
Beyond,'' while stressing the cost and difficulty of achieving
accurate counts in the hardest to enumerate places, admitted in
his testimony that, ``[o]n the other hand, the use of surveys
and sampling techniques will mean that the estimates for very
small areas will have greater variation above and below the
true count.'' \51\
---------------------------------------------------------------------------
\51\ Charles Schultze, testimony before the Committee on Government
Reform and Oversight, February 29, 1996.
---------------------------------------------------------------------------
Dr. Schultze's testimony centered around the findings of
the Census 2000 panel, which produced a book entitled
``Modernizing the U.S. Census'' (1995, National Academy of
Sciences). According to the findings of the panel, data
accuracy is relative to the particular size of the geographical
or political area.\52\ In other words, a census incorporating
nonresponse sampling and an adjustment based on the ICM sample
could raise the level of accuracy for a numeric estimate at the
national level (numeric accuracy). However, the degree of error
introduced by those techniques increases as the estimates are
applied to the State and then sub-state levels (distributive
accuracy). This has implications for congressional and State
legislative redistricting, as well as distribution of Federal
funds at the State level and below.
---------------------------------------------------------------------------
\52\ Modernizing the U.S. Census, National Academy of Sciences,
National Academy Press, 1995, p. 35.
---------------------------------------------------------------------------
Ultimately sampling, be it for nonresponse or ICM, does not
allow for knowing definitively the individual attributes of the
population it is imputing into an area. In particular, sampling
does not allow for knowing where the people missed actually
live, so the accuracy of the population distribution becomes
inferior. For congressional redistricting and for local-area
decisions that involve smaller areas, the relative accuracy of
the population count for blocks and aggregations of blocks is
very important. In sum, the committee found overwhelming
concern among Members of Congress and outside experts that
sampling would compromise the quality of data at the small, but
critical, geographic areas such as blocks and aggregations of
blocks.
Again, while driven toward nonresponse sampling primarily
by cost concerns, it appears that the fundamental
constitutional purpose for the census--to apportion Congress--
has been de-emphasized. The Department of Commerce and the
Bureau of the Census should prioritize the constitutional basis
of the decennial census as well as their use for congressional
and State legislative redistricting. The Bureau should strive
hardest for accuracy and fairness in getting the proportional
distribution of the population physically right among
geographical and political units in order to fulfill our
constitutional mandate, first and foremost.
6. The complexity of the two different sampling techniques being
planned for the 2000 Census adds a great deal of risk to the
operational feasibility of the Bureau's current approach
In 1990, in the attempt to adjust the census, the Bureau
sampled 150,000 households in what it called a ``post
enumeration survey,'' or PES. The results of the survey were
matched to the original count to adjust for the estimated
undercount/overcount. This was one of the largest surveys ever
undertaken, and it was not free from error, as indicated below.
For the 2000 Census, the Bureau plans to sample 750,000
households, or five times the number of households in the 1990
sample. The committee has concern about the sheer volume of
this sample and the capability of the Census Bureau to
operationally carry out a sample of this magnitude and
complexity in the time to meet the statutory deadline for
completion of the census. The Members are concerned as well
that there are no plans for independent analyses of the survey
to measure its quality prior to its use in adjusting the
census.
Dr. Wachter, who testified as an expert witness for the
Government in the 1980 and the 1990 lawsuits over census
adjustment, expressed his concern at the February hearing about
the complexity of the plans for the statistical adjustments to
the 2000 Census:
The first priority should be a `Fail-Safe Census'.
Statistical methods need to be simple and direct so
that malfunctions will be detected and corrected.
Unfortunately, the Bureau's plans for Census 2000 add
further layers of complexity onto the complications of
1990, and leave the final numbers even more vulnerable
to statistical error.\53\
---------------------------------------------------------------------------
\53\ Wachter, testimony, February 29, 1996.
Indeed, with either statistical methodology contemplated
for ICM, CensusPlus or DSE, the potential for added error
appears great when one considers the numerous sources of error.
In the Census Bureau memo detailing the results of the 1995
Census Test (Memorandum No. 46, April 1, 1996), a host of error
sources are evaluated, and include: (1) contamination error,
(2) reconciliation bias, (3) outmover error, (4) rostering
error, (5) combined data collection error, (6) CensusPlus
residency coding error, (7) DSE matching error, (8) missing
data, and (9) sampling error.\54\
---------------------------------------------------------------------------
\54\ ``1995 Census Test Results,'' Memorandum No. 46, by E. Ann
Vacca, Mary Mulry, Ruth Ann Killion, Decennial Statistical Studies
Division, April 1, 1996, p. 28.
---------------------------------------------------------------------------
The importance of assessing the quality of a survey, or
sample, cannot be overstated. The General Accounting Office, in
its 1991 report entitled ``The 1990 Census Adjustment,''
indicates that:
The difficulties in successfully completing the PES
and the fact that the PES, like all surveys, is subject
to a variety of errors, underscores the importance of
completing sound and careful assessments of the quality
of the PES. . . . The value of the PES estimates will
be appreciably reduced if the error in those estimates
is considered significant. Thus assessments of the PES
form a critical part of the data the Secretary of
Commerce will need to make the adjustment decision.\55\
---------------------------------------------------------------------------
\55\ ``1990 Census Adjustment,'' General Accounting Office, GAO/
GGD-91-42, March 1991, p. 16.
Yet, under the Bureau's concept for a ``one-number census''
there will not be a set of figures for the actual enumeration
(completed with or without the sample for nonresponse) against
which to independently evaluate the ICM estimate. Only the
final ICM version will be made available to the public. Dr.
Wachter, in his testimony before the committee, also expressed
concerned about this issue. ``The figures behind the final
figures will not be available outside the Bureau. If choices of
detail shift a dozen seats in the House of Representatives, we
shall never know. If the problems of 1990 are brought under
control by the Bureau's new initiatives, we shall never know.''
\56\
---------------------------------------------------------------------------
\56\ Wachter, testimony, February 29, 1996.
---------------------------------------------------------------------------
Bureau's Plans Limit Quality Control
In addition, the Bureau, citing funding constraints, plans
only a limited quality control program compared to the 1990
program where quality control was conducted for more operations
and with higher sampling rates.\57\ For the 2000 Census, the
Bureau will implement quality control operations only to detect
major errors in data collection and processing operations. The
most serious implication of this plan is stated by the Bureau:
``Data collection and processing operations may yield results
of lower quality than those achieved in the 1990 Census.'' \58\
---------------------------------------------------------------------------
\57\ The Plan, p. III-39.
\58\ Id., p. III-39.
---------------------------------------------------------------------------
Even with the quality control procedures in place for 1990,
the committee was reminded of a major problem that arose with
the 1990 attempted adjustment. Senator Kohl stated at the
hearing in February that, ``[i]n retrospect, we were extremely
fortunate that we did not adjust the 1990 Census. Several
months after that contentious decision in July of 1991, the
Census Bureau discovered an error in the adjustment procedures
that significantly reduced the undercount.'' \59\
---------------------------------------------------------------------------
\59\ Senator Herb Kohl, testimony before the House Government
Reform and Oversight Committee, February 29, 1996.
---------------------------------------------------------------------------
Dr. Wachter echoed this concern about potential errors. He
recalled that, ``If Secretary Mosbacher had decided to use the
statistically adjusted numbers as 1990 census counts, a seat in
the House of Representatives would have been shifted from
Pennsylvania to Arizona by an error in a computer program. It
affected a million people in the count. It remained
undiscovered for months after the Secretary's decision, buried
under layer upon layer of complications in the statistical
procedures.'' \60\
---------------------------------------------------------------------------
\60\ Wachter, testimony, February 29, 1996.
---------------------------------------------------------------------------
The computer coding error, according to Dr. Wachter:
epitomizes the problems of statistical adjustment in
1990. Complications were added in pursuit of
incremental gains in accuracy. However, many small
gains can be offset by a few large errors. In 1990, the
complexity of their modeling systems made it hard for
the Bureau to detect big mistakes and uncertainties,
until long after the critical decisions had been made.
Indeed, we and others inside and outside the Bureau
later found systematic errors in the procedures
originally used to evaluate the adjustment proposed to
Secretary Mosbacher: the original evaluations
overstated the merits of adjustment. The more complex
are the modeling systems used, and the tighter the
constraints of time and money, the harder it becomes to
make realistic assessments of the statistical
uncertainties in the model outputs.\61\
---------------------------------------------------------------------------
\61\ Id.
Dr. Barbara Bailar, former Associate Director for
Statistical Standards and Methodology at the Census Bureau,
testified about her concerns for the ability to carry out
---------------------------------------------------------------------------
quality control with the complex plans for the 2000 Census:
I am concerned that the Bureau is instituting so many
new procedures and processing steps without having
adequate quality control. Not that I am advocating
quality control for documentation purposes only, but I
believe procedures should be built into the census
processes that tell the Census staff quickly if
something is going wrong. That the Bureau staff
worrries that funding constraints limit their quality
control program makes me very concerned.\62\
---------------------------------------------------------------------------
\62\ Barbara Bailar, testimony before the House Government Reform
and Oversight Committee, February 29, 1996.
---------------------------------------------------------------------------
IV. Recommendations
1. Congress should work to clarify existing Federal statutes with
regard to the use of sampling to make statistical adjustments
to the census for apportionment purposes
The reason there are so many lawsuits over the issue of
sampling/adjustment is because Federal statutes have been
interpreted in various ways. It is the desire of this committee
to avoid the numerous lawsuits that came after Secretary
Mosbacher's decision not to adjust the census. Litigation is
expensive for the American taxpayer. Though we do not have
exact figures, we expect that it has cost plenty for the
Federal Government to argue its cases before the courts.
Congress should clarify its intent with respect to sampling/
adjustment by amending 13 U.S.C., sections 141 and/or 195.
2. The Bureau should not use sampling methods to complete or adjust the
actual enumeration of the 2000 Census which is constitutionally
mandated for purposes of apportionment
Congress should not allow sampling/statistical adjustment
to modify or replace the actual enumeration. The primary reason
is that a statistical adjustment is inherently subjective and
open to potential manipulation of the final count. It is also
possible that sampling/adjustment could undermine public
confidence in the census and erode participation by State and
local groups, and individuals.
3. The Department of Commerce and the Bureau of the Census should
prioritize the constitutional mandate of the decennial census--
apportionment of the House of Representatives
It appears that the fundamental constitutional purpose for
the decennial census--to apportion the House of
Representative--has been de-emphasized. However, this is the
element that must take precedence--regardless of cost. The
Bureau should strive hardest for accuracy and fairness in terms
of getting the proportional distribution of the population
physically right among geographical and political units. The
committee believes this can only be achieved by performing an
actual enumeration.
4. The Bureau should emphasize and strengthen its cooperative
relationships with State and local elected officials, as well
as members of local organizations, who are vital in helping
increase response rates to the decennial census
The committee recognizes and is concerned that there has
always been an undercount, and that in the 1990 Census there
was an increase in the differential undercount of minorities.
This is not the fault of the Bureau of the Census. Socio-
economic changes have impacted greatly on the ability of the
Bureau to locate everyone. Additionally, there are more people
who are skeptical of ``government'' and who refuse to cooperate
with the census. These people exist in all areas of this
country, urban, rural, and suburban. However, no-one is
omniscient. We do not know for sure exactly what these people
look like. We do not know, nor have we ever known, the ``true''
population.
The Bureau is spending considerable time and resources
developing ways to improve response rates and encourage
participation in the census. A major effort will be in
encouraging stronger relationships with State and local elected
officials, as well as nongovernmental organizations, who can
help educate communities about the importance of the census to
individuals personally. They know their communities better than
anyone. And it is clearly in their interest to make sure
everyone is counted. To embark on a program of estimating the
last 10 percent of the population would certainly undermine
this important effort.
5. The Bureau should strengthen its plans for a thorough quality check
of the 2000 Census and maintain open access to all processes
for internal and external review and analysis
It is disturbing that the Bureau is so dramatically de-
prioritizing quality control in the 2000 Census. In doing so,
not only is the Bureau jeopardizing the quality of the
decennial census, but it is jeopardizing its reputation as the
finest statistical agency in the world. Resources must be made
available to evaluate the complicated steps in an undertaking
of this magnitude. This country can ill-afford a $4 billion
mistake.
CLARIFYING COMMENTS OF HON. WILLIAM F. CLINGER, JR.
There are many subjective elements in the adjustment
process, including choices of adjustment methodology,
statistical models for estimation, post stratification, and
sample design. Experience in 1990 showed these choices had
substantial impacts, and introduced undesirable, arbitrary
elements into the census process. However, with the probability
sampling techniques used by the Census Bureau, given the sample
design, the choice of the sample itself is objective.
Hon. William F. Clinger, Jr.
ADDITIONAL VIEWS OF HON. STEVEN SCHIFF
The decennial census and how it is conducted have always
been sensitive and controversial matters. The data from the
final census are used for a variety of purposes. The
Constitutional mandate to apportion the U.S. Congress is by far
the most recognized purpose of the Census. But, census data are
also used in the allocation of billions of dollars each year in
Federal and State grants.
The Census Bureau has decided for the first time to use
sampling techniques in the 2000 decennial census to adjust the
actual enumeration. With the stakes being so high for States
and localities, this is undoubtedly a very important decision.
I agree with many of the recommendations in the committee's
report. I especially agree with the recommendation that the
Bureau should enhance partnerships with State and local
officials in order to provide the most accurate survey
possible. This would be most helpful for those areas, such as
New Mexico, which have a significant percentage of the
population who for a variety of reasons do not wish to respond
to the initial mail-out form. In my view, the Bureau should
make every effort in conjunction with State and local
governments to account for nonrespondents by physical means.
All reasonable efforts should continue to be made to
achieve a maximum actual count of individuals. However, after
such count is complete, I believe that there is still
additional room for finalizing population numbers through the
use of sampling.
No one believes that any process of individual
identification will produce a totally accurate result. Using
reasonable sampling techniques, after a detailed count, will
help us achieve a more accurate final result, than by an
individual count, alone.
Hon. Steven Schiff.
ADDITIONAL VIEWS OF HON. ILEANA ROS-LEHTINEN
I have several comments about the committee report:
``Sampling and Statistical Adjustment in the Decennial Census:
Fundamental Flaws'' which I am stating here for the record.
The report rejects the use of statistical adjustment to
help address the undercount problem in the 1990 census, and in
particular, the undercount of minority population. I believe
that a statistical adjustment must be looked at as a possible
way to improve the census count, but that we must ensure that
accuracy is not sacrificed in any adjustment for an undercount.
I am deeply concerned about a possible undercount in the
next census in 2000 which would unfairly impact on my home
State of Florida and other areas throughout the Nation with
high growth rates and high minority populations. The Census
Bureau estimates that the last census failed to count 340,000
people in Florida or approximately 2.6 percent of the
population and over 60,000 people or 3.0 percent of the
population in Metropolitan Dade County. This compares with a
2.1 percent undercount of the total U.S. population at the
national level, and an undercount of 4.8 percent of the black
population and 5.2 percent of the Hispanic population
throughout the Nation.
This undercount costs Florida and other areas untold
millions of dollars in Federal funds. It is essential that the
2000 census be as accurate as possible since it determines such
important matters as the size of each State's Congressional
delegation, its votes for President in the Electoral College,
the boundaries of numerous districts at all levels of
government, and the determines the distribution of government
funding formulas.
I believe that the Census Bureau's proposed plan for
sampling for nonresponse in the 2000 census needs further
modification, especially in ensuring accuracy for small area
population figures which are important in assuring fairness in
redistricting based on the census figures. I have joined with
my Miami colleague, Rep. Carrie Meek, in co-sponsoring H.R.
3558 which would require the Census Bureau to count 90 percent
of the population at the census tract level before sampling.
This approach of sampling at the census tract level is
supported by Chuck Blowers, Chief of Research and Oliver Kerr
of the Metropolitan Dade County Department of Planning,
Development and Regulation. They agree that this requirement
``will result in more accurate census counts, especially in
areas of minority concentration and those with many recent
immigrants.''
I strongly concur with the report's recommendation that the
Census Bureau ``should emphasize and strengthen its cooperative
relationships with state and local elected officials, as well
as members of local organizations, who are vital in helping
increase response rates to the decennial census''. My local
county planning officials would appreciate having access to the
Census Bureau's address listings by tract and block far enough
in advance so that they can make corrections to it prior to the
next census.
Hon. Ileana Ros-Lehtinen.
DISSENTING VIEWS OF HON. CARDISS COLLINS, HON. HENRY A. WAXMAN, HON.
TOM LANTOS, HON. MAJOR R. OWENS, HON. EDOLPHUS TOWNS, HON. JOHN M.
SPRATT, JR., HON. LOUISE MCINTOSH SLAUGHTER, HON. GARY A. CONDIT, HON.
BERNARD SANDERS, HON. KAREN L. THURMAN, HON. CAROLYN B. MALONEY, HON.
BARBARA-ROSE COLLINS, HON. ELEANOR HOLMES NORTON, HON. JAMES P. MORAN,
HON. GENE GREEN, HON. CARRIE P. MEEK, HON. CHAKA FATTAH, AND HON.
ELIJAH E. CUMMINGS
The majority has proposed a report that rejects sampling to
count the most difficult to count in the census, and adjusting
the census for those that cannot be found. Yet both have been
proposed by the Census Bureau to constrain the cost of the
census and make it more accurate. This outright rejection of
sampling and adjustment, without any proposal for achieving the
dual charge from Congress of a more accurate and less expensive
census, is untenable.
There has been considerable confusion about the proposals
to use sampling in the 2000 census. The Census Bureau has
proposed two uses of sampling: once 90 percent of the
households have been counted, the Census will draw a 1 in 10
sample of the remaining households (commonly referred to as
``sampling for nonresponse''); and second, a separate survey
will be conducted to determine who was missed or counted twice,
and the population counts adjusted accordingly (called
Integrated Coverage Measurement by the Census Bureau). Several
members have expressed concerns about the implementation of
sampling for nonresponse, for example, Rep. Meek has introduced
H.R. 3558 to constrain the geography used for this sample.
Passing this report puts the committee on record opposing
any adjustment. If the 2000 census is not adjusted for the
undercount of minorities, it would continue a 60 year tradition
of measuring the problem, but doing nothing about it. Congress
has repeatedly called for a more accurate census. With the
passage of this report, we will be settling for an inaccurate
census.
Dr. Barbara Bryant, Director of the Census Bureau under
President Bush, in testimony before the House Committee on Post
Office and Civil Service, said that the census had reached the
limits of what could be done with traditional methods. Congress
has called for a census that is less expensive and more
accurate.\1\ Three separate panels of experts convened by the
National Academy of Sciences have recommended the use of
sampling and statistical methods to achieve these goals.
---------------------------------------------------------------------------
\1\ Dr. Barbara Bryant, testimony before the House Committee on
Post Office and Civil Service, July 16, 1991.
---------------------------------------------------------------------------
We agree that the plans laid out by the Census Bureau need
further definition, and that the plans for sampling for
nonresponse need to be modified. Where we differ is the
resolution. The Republican report offers no solutions. We must
be sure that methods used to improve accuracy and reduce costs
are, as Dr. Kenneth Wachter testified before this committee,
``simple, direct, and fail-safe.'' However, to prohibit new
methods that could reduce cost or improve accuracy is
premature.
findings
1.The fundamental purpose of the decennial census of
population is to account for all residents of the United States
for the purpose of apportioning the seats in the House of
Representatives among the States.
2. Sixty years of research on the census has shown that
there will always be some residents for whom the census has not
accounted. The net undercount has come down from 5.4 percent in
1940 to 1.2 percent in 1980, but rose to 1.8 percent in
1990,\2\ suggesting that we may have reached the limit to which
we can expect the census to go using traditional direct
enumeration methods.
---------------------------------------------------------------------------
\2\ Hogan, Howard and Gregg Robinson, ``What the Census Bureau's
Coverage Evaluation Programs Tell Us About Differential Undercount,''
unpublished paper.
---------------------------------------------------------------------------
3. It is unlikely that the census can account for the final
1 to 2 percent of the population without employing some type of
statistical procedure. In order for these methods to achieve
widespread public acceptance, they must be simple. The results
cannot be altered by changing the assumptions within the
method, and they must be subject to widespread independent
review by both the public and the professional community.
4. The cost of a traditional census has increased
dramatically, doubling in constant dollars from 1970 to 1980,
and increasing another 25 percent in 1990. Part of the cost
increase is the result of a declining percentage of forms
returned by mail, from 78 percent in 1970 to 75 percent in 1980
to 65 percent in 1990; however; the largest increase in cost
does not correspond with the largest drop in the mail-back
response rate.\3\
---------------------------------------------------------------------------
\3\ Decennial Census: 1990 Results Show Need for Fundamental
Reform, U.S. General Accounting Report GAO/GGD-92-94, June 1992. Other
factors for declining mail returns, cited by the General Accounting
Office, include illiteracy, non-English-speaking immigrants, concerns
about privacy, the hectic nature of modern living, undocumented aliens,
growth in commercial mail and telephone solicitations, lack of
confidence in civic institutions, and increasing numbers of
nontraditional households and family arrangements.
---------------------------------------------------------------------------
5. The Census Bureau has proposed a budget of $3.9 billion
for the 2000 census,\4\ with major savings achieved by
accounting for the last 10 percent of the population through a
1 in 10 sample. However, Congress has shown a reluctance to
fund the census at this level. Both the FY 1995 and FY 1996
budget resolutions funded below the requested level, and the
chairman of the Commerce, State, and Justice Appropriations
Subcommittee has indicated that the Census Bureau will not be
funded at the requested level for FY 1997.\5\ Both the House
and Senate Appropriations subcommittees have proposed funding
only about two-thirds of the increase requested to fund 2000
census activities in FY 1997.
---------------------------------------------------------------------------
\4\ The Plan for Census 2000, Bureau of the Census, U.S. Department
of Commerce, April 1996.
\5\ Roll Call, May 6, 1996.
---------------------------------------------------------------------------
6. The implementation of sampling to account for the last
10 percent has not received wide-spread public acceptance.
Members of Congress, as well as the Census Bureau's African-
American Advisory Committee,\6\ have expressed strong
reservations about the choice of geography and detrimental
effects on the count of minorities. On the other hand, a
National Academy of Sciences panel,\7\ convened at the request
of the Census Bureau, strongly supported sampling to account
for the last 10 percent. In testimony before the Government
Reform and Oversight Subcommittee on National Security,
International Affairs, and Criminal Justice, both the General
Accounting Office \8\ and the Department of Commerce Inspector
General \9\ supported sampling.
---------------------------------------------------------------------------
\6\ ``Minutes of the December 1995 Meeting of the Minority Advisory
Committees'', U.S. Bureau of the Census, unpublished.
\7\ White, Andrew A. And Keith F. Rust, eds., Sampling in the 2000
Census: Interim Report I, Panel to Evaluate Alternative Census
Methodologies, Committee on National Statistics, National Research
Council, National Academy Press, Washington, DC, June 1996.
\8\ L. Nye Stevens, Director of Federal Management and Workforce
Issues, U.S. General Accounting Office. Testimony before the House
Government Reform and Oversight Subcommittee on National Security,
International Affairs, and Criminal Justice, October 25, 1995.
\9\ Francis DeGeorge, Inspector General, U.S. Department of
Commerce. Testimony before the House Government Reform and Oversight
Subcommittee on National Security, International Affairs, and Criminal
Justice, October 25, 1995.
---------------------------------------------------------------------------
background
Finding 1. The fundamental purpose of the decennial census of
population is to account for all residents of the United States
for the purpose of apportioning the seats in the House of
Representatives among the States.
The decennial census of population was created in Article
I, Section 2 of the Constitution:
Representatives and direct Taxes shall be apportioned
among the several States which may be included in the
Union, according to their respective Numbers, . . . The
actual enumeration shall be made within three Years
after the first Meeting of the Congress of the United
States, and within every subsequent Term of ten Years,
in such Manner as they shall by Law direct.\10\
---------------------------------------------------------------------------
\10\ The Constitution of the United States, Article I, Section 2.
The Constitution makes no mention of any criteria for being
included in the census other than residence. Both the courts
and Congress have affirmed that the census is to include all
residents. In 1990, the U.S. District Court in Pittsburgh, PA
threw out a lawsuit that would have blocked the Census Bureau
from counting illegal aliens in the 1990 census. A similar suit
was dismissed in 1980.\11\ In August 1989, Congress rejected an
amendment to the FY 1990 appropriations for the Department of
Commerce that would have required the Census Bureau to exclude
illegal aliens from the 1990 census count.\12\ Rep. Stephen
Horn expressed these sentiments at the February 29, 1996
hearing of the Committee on Government Reform and Oversight,
``. . . as you know, the Constitution says nothing about
citizens; it's `persons.' '' \13\
---------------------------------------------------------------------------
\11\ Rich, Spencer, ``Suit to Block Illegal Aliens From Census
Count Voided,'' the Washington Post, May 10, 1989, p. A 17.
\12\ Biskupic, Joan, ``Dodging Touchy Census Issue, House Passes
Funding Bill,'' Congressional Quarterly, August 5, 1989, p. 2047.
\13\ Transcript of the Committee on Government Reform and Oversight
hearing ``The 2000 Census: Putting Our Money Where it Counts,''
February 29, 1996, p. 48.
---------------------------------------------------------------------------
In Franklin v. Massachusetts \14\ the Supreme Court
rejected a claim by the State of Massachusetts and two of its
registered voters. Massachusetts argued that the method used by
the Census Bureau to count Federal employees serving overseas
was consistent with the constitutional language and goal of
equal representation. The Supreme Court upheld the authority of
the Secretary of Commerce to define the boundaries of ``usual
residence''.
---------------------------------------------------------------------------
\14\ 505 U.S. 788 (1992).
---------------------------------------------------------------------------
Finding 2. Sixty years of research on the census has shown that there
will always be some residents for whom the census has not
accounted. The net undercount has came down from 5.4 percent in
1940 to 1.2 percent in 1980, but rose to 1.8 percent in
1990,\15\ suggesting that we may have reached the limit to
which we can expect the census to go.
---------------------------------------------------------------------------
\15\ Hogan, Howard and Gregg Robinson, ``What the Census Bureau's
Coverage Evaluation Programs Tell Us About Differential Undercount,''
unpublished paper.
---------------------------------------------------------------------------
There has been marked improvement in reducing the
undercount since the first estimates from a Census Bureau
program called Demographic analysis. Demographic Analysis is
limited because it only gives measures for the whole country,
and cannot provide estimates for ethnic groups or racial groups
other African-American. However, it has a long history of
research, provides more age detail than other measures, and
provides greater confidence in describing differences in the
undercount between groups.\16\ The table below gives the net
undercount for the United States from 1940 to 1990.\17\
---------------------------------------------------------------------------
\16\ Ibid. p. 5.
\17\ Ibid. p. 5.
----------------------------------------------------------------------------------------------------------------
1940 1950 1960 1970 1980 1990
----------------------------------------------------------------------------------------------------------------
Total..................................................... 5.4% 4.1% 3.1% 2.7% 1.2% 1.8%
Black..................................................... 8.4% 7.5% 6.6% 6.5% 4.5% 5.7%
Nonblack.................................................. 5.0% 3.8% 2.7% 2.2% 0.8% 1.3%
Difference (Black - Nonblack)............................. 3.4% 3.6% 3.9% 4.3% 3.7% 4.4%
----------------------------------------------------------------------------------------------------------------
While the net undercount for the United States has declined,
the black/nonblack differential has increased with each census,
with the exception of 1980, reaching the highest measured
difference in 1990. This differential draws into question the
equity of the census data, especially when used for
apportionment, and by extension the electoral college, as well
as for distributing Federal funds. It shows clearly that the
census does not account for all residents of the United States.
Over 4 million people were left out of the 1990 census.\18\
---------------------------------------------------------------------------
\18\ The undercount for the census is always reported as a net
figure which includes both people counted more than once, and people
not counted at all. In 1990 approximately 10 million people were missed
in the census and 6 million people were counted twice, resulting in a
net undercount of 4 million.
---------------------------------------------------------------------------
In addition to racial differentials, the undercount varies
by geography. Following the 1990 census, the Census Bureau
conducted a ``post-enumeration survey'' which provides more
geographic detail than demographic analysis as well as
estimates for other racial and ethnic groups.
The post-enumeration survey shows considerable differences
in the net undercount from State to State, from a low of 0.3
percent in Pennsylvania to a high of 3.4 in the District of
Columbia. The undercount is often talked about in terms of the
problems of enumeration in urban areas and the problems of high
undercounts for minorities. However, many non-urban States with
a larger than average rural population have undercount rates
above the national average of 1.6 percent:\19\ Idaho (2.0
percent); Montana (2.4 percent); Mississippi (2.1 percent);
North Carolina (1.8 percent); and South Carolina (2.0 percent).
Large undercounts also occur in States like California (2.7
percent), Florida (2.0 percent); Louisiana (2.2 percent),
Maryland (2.1 percent); New Mexico (3.1 percent) and Texas (2.8
percent).\20\
---------------------------------------------------------------------------
\19\ The estimate of a 1.6 percent net undercount is from the post-
enumeration survey. The estimate of 1.8 percent is from demographic
analysis. The two methods yield slightly different estimates for the
same group, but there are no distinct differences.
\20\ Hogan, Howard, ``The 1990 Post-Enumeration Survey: Operations
and New Estimates,'' paper presented at the 1992 Annual Meetings of the
American Statistical Association, August 1992, U.S. Bureau of the
Census, Washington, DC, and unpublished tables provided by the Bureau
of the Census.
---------------------------------------------------------------------------
Finding 3. It is unlikely that the census can account for the final 1
to 2 percent of the population without employing some type of
statistical procedure. In order for these methods to achieve
widespread public acceptance, they must be simple. The results
cannot be altered by changing the assumptions within the
method, and they must be subject to widespread independent
review by both the public and the professional community.
The demands of a statistical adjustment are severe. The
census counts people block by block across the country, and
those counts by block are used to build congressional
districts, State legislature districts, and to allocate funds
for a variety of Federal programs like the grants to school
districts for disadvantaged children under Title I of the
Elementary and Secondary Education Act.
Debate over the proposed adjustment for the 1990 census was
fierce. Dr. Barbara Bryant, Director of the Census Bureau
during the 1990 census, acknowledged that the procedures were
ones over which reasonable demographers and statisticians could
disagree. Some, like the State of Wisconsin, argued that
adjusting the census would be unfair to those States that made
an effort to get a complete count. Others argued for the
fairness of an adjustment. Peter Chacon, then chairman of the
California State Assembly Elections and Reapportionment
Committee testified ``I would close my testimony by urging the
Subcommittee to do all in its power to impress upon the
Secretary of Commerce that he should order an appropriate
adjustment of the 1990 census so that undercounted minorities
can achieve the political representation to which they are
entitled.'' \21\
---------------------------------------------------------------------------
\21\ Chacon, Peter, testimony at the joint hearing before the
Subcommittee on Census and Population of the House Committee on Post
Office and Civil Service and the Subcommittee on Government Information
and Regulation of the Senate Committee on Governmental Affairs, March
19, 1991, Serial No. 102-5.
---------------------------------------------------------------------------
The fundamental uses of census data require that any shift
from the traditional methods be done with a clear focus on
public perception. In his testimony before the Committee on
Government Reform and Oversight, Rep. Tom Sawyer said ``We are
talking about accuracy; we are talking about cost; and we are
talking about confidence.'' In answering questions he
elaborated, ``I think we need to be able to adjust the
techniques that we use, but we should do so in a way that
sustains the broad public confidence and improves the
accuracy.'' \22\ Dr. Kenneth Wachter echoed this sentiment when
he said ``Statistical methods need to be simple and direct so
that malfunctions will be detected and corrected.'' He went on
to state, ``. . . if statistical methods are to be given a role
in the generation of the Census 2000 counts, then those methods
should be simple, direct, and fail-safe.'' \23\
---------------------------------------------------------------------------
\22\ Transcript of the Committee on Government Reform and Oversight
hearing ``The 2000 Census: Putting Our Money Where it Counts,''
February 29, 1996, p. 21 and p. 44.
\23\ Ibid., p. 117.
---------------------------------------------------------------------------
Finding 4. The cost of the census has increased dramatically, doubling
in constant dollars from 1970 to 1980, and increasing another
25 percent in 1990. Part of the cost increase is the result of
a declining rate of forms returned by mail, from 78 percent in
1970 to 75 percent in 1980 to 65 percent in 1990, however; the
largest increase in cost does not correspond with the largest
drop in the mail-back response rate.
Between 1984 and 1993, the Census Bureau spent $2.6 billion
to plan, conduct, process, and publish the 1990 census.\24\ The
Census Bureau and GAO estimate that the same census in 2000
would cost $4.8 to $5 billion.\25\ The 1970 census cost $10 per
household. In 1980, the cost per household, in constant
dollars, was $20 per household, and in 1990, the census cost
$25 per household. The $3.9 billion estimate would hold the
cost in constant dollars to about $25 per household.
---------------------------------------------------------------------------
\24\ According to Census Bureau figures, data collection accounted
for 49.7 percent of the total cost. Data processing and geographic
support accounted for an additional 28.6 percent. The remaining 21.7
percent was for planning and direction (6.6 percent), data
dissemination (5.8 percent), research and evaluation (4.4 percent),
promotion and outreach (2.8 percent), and other activities (2.1
percent).
\25\ The Plan for Census 2000, Bureau of the Census, U.S.
Department of Commerce, April 1996.
---------------------------------------------------------------------------
At the same time costs have been escalating, participation
in the mail-out/mail-back portion of the census has been
declining. In 1970 the Census Bureau began to use the U.S.
Postal Service to deliver census forms. During that census, 78
percent returned the form by mail. In 1980, 75 percent of the
households returned the form by mail. For the 1990 census, the
Census Bureau budgeted for a 70 percent mail-back return rate,
however, only 65 percent were return by mail. Current research
by the Census Bureau suggests that the comparable rate for 2000
will be approximately 55 percent. Response rates also vary by
geography. Large metropolitan areas usually experience the
lowest mail-back rates, and 23 of the 32 largest cities had
response rates lower than the national average. Boston,
Chicago, Cleveland, New Orleans, New York, and Washington, DC,
all had mail-back response rates below 55 percent.
The decline in participation, particularly in returning the
form by mail, is often identified as a major source of the
increased cost of the census. The National Academy of Sciences
panel on census requirements stated ``A substantial decline in
the population's response rate to the mailed census
questionnaire . . . has been an important cause of the cost
escalation, . . .'' \26\ Similarly, Dr. Everett Ehrlich, Under
Secretary for Economic Affairs at the Department of Commerce,
at the June 6, 1996 hearing before the Committee on Government
Reform and Oversight, said ``The cost of the census per
respondent in 1990 dollars went up very dramatically,
particularly, in 1990, up to 25 1990 dollars per respondent,
and it did so because of this fundamental failure of design
that I discussed earlier. That is, participation rates were
falling, . . .'' \27\
---------------------------------------------------------------------------
\26\ Edmonston, Barry and Charles Schultze, eds., Modernizing the
U.S. Census, Panel on Census Requirements in the Year 2000 and Beyond,
Committee on National Statistics, National Research Council, National
Academy Press, 1995, p. 44.
\27\ Dr. Everett Ehrlich, transcript of testimony before the
Committee on Government Reform and Oversight, June 6, 1996, p. 72.
---------------------------------------------------------------------------
The doubling of cost between 1970 and 1980 corresponds with
a 3 percentage point decrease in the mail-back response rate.
Between 1980 and 1990, the mail-back response rate fell 10
percentage points, but corresponded to a 25 percent increase in
cost. The National Academy of Sciences panel suggests that this
anomaly may be due to a more intensive effort to enumerate
everyone through labor-intensive measures, a decline in the
quality of temporary personnel, and an increased demand for
accurate counts at small geographic levels.\28\
---------------------------------------------------------------------------
\28\ Op. Cit. p. 53.
---------------------------------------------------------------------------
Finding 5. The Census Bureau has proposed a budget of $3.9 billion for
the 2000 census, with major savings achieved by accounting for
the last 10 percent of the population through a 1 in 10 sample.
However, Congress has shown a reluctance to fund the census at
this level. Both the FY 1995 and FY 1996 budgets were funded
below the requested level, and the chairman of the Commerce,
State, and Justice Appropriations Subcommittee has indicated
that the Census Bureau will not be funded at the requested
level for FY 1997.
On February 28, 1996, the Census Bureau announced its plans
for the 2000 census. Innovations to past censuses include a
redesigned ``user-friendly form; plans to mail reminder letters
and replacement forms to those who do not return their form by
mail; a plan to account for the last 10 percent of the
population with a sample (sampling for nonresponse); and an
adjustment procedure incorporated into the census called
``integrated coverage measurement.'' \29\ The estimated cost
for the 2000 census is $3.9 billion; just under the $25 per
household cost of the 1990 census. Sampling for nonresponse
accounts for about $500 million of the $900 million savings
over the estimated $4.8 billion cost of repeating the
procedures from 1990.
---------------------------------------------------------------------------
\29\ The Plan for Census 2000, Bureau of the Census, U.S.
Department of Commerce, April 1996.
---------------------------------------------------------------------------
Congress has shown a reluctance to fund the 2000 census
even at the $3.9 billion estimated by the Census Bureau. In FY
1994, the first year of the 2000 census budget cycle, the
Census Bureau requested $23.1 million for census 2000
activities, and was allocated $18.7 million by the
Appropriations Subcommittee on Commerce, Justice, State and
Judiciary. In FY 1995 the Census Bureau requested 48.6 and
received $42.1 million. The Census 2000 request for FY 1996 was
$60.1 million, and the appropriation was $50.6 million. In FY
1997, the Census Bureau has requested a total of $105 million
for the 2000 census activities. Following the Appropriations
Subcommittee hearing on the Census Bureau budget, Chairman Hal
Rogers was quoted as saying that the $3.9 billion is ``entirely
too much money'' and said that he was ``disinclined to agree to
the funding request'' of $105.\30\ Both the House and Senate
Appropriations Subcommittees have proposed funding only about
two-thirds of the increase requested to fund 2000 census
activities in FY 1997. At the June 6, 1996, Committee on
Government Reform and Oversight hearing, Rep. Gene Green said,
``. . . you cannot have it both ways. If we are going to cut
the budget for the census, we are going to have to sample more.
. . .'' \31\
---------------------------------------------------------------------------
\30\ Roll Call, May 6, 1996.
\31\ Transcript of testimony before the Committee on Government
Reform and Oversight, June 6, 1996, p. 25.
---------------------------------------------------------------------------
Finding 6. The implementation of sampling to account for the last 10
percent has not received wide-spread public acceptance. Members
of Congress, as well as the Census Bureau's African-American
Advisory Committee, have expressed strong reservations about
the choice of geography and detrimental effects on the count of
minorities. On the other hand, a National Academy of Sciences
panel, convened at the request of the Census Bureau, strongly
supported sampling to account for the last 10 percent. In
testimony before the Government Reform and Oversight
Subcommittee on National Security, International Affairs, and
Criminal Justice, both the General Accounting Office and the
Department of Commerce Inspector General supported sampling.
The implementation of sampling to account for the last 10
percent has not received wide-spread public acceptance. At the
February 28, 1996, announcement of the 2000 plan, the Census
Bureau reported that following the second questionnaire
mailing, it would either phone or visit households that had not
mailed back the form, until it had counted 90 percent of each
county. At the December 1995 meeting of the minority advisory
committees, Dr. Robert Hill, a member of the African-American
Advisory Committee, commented that he ``had not seen evidence
that the Bureau would do extra outreach activities in minority
communities; distributing paper advertisements will not
suffice.'' \32\ Ms. Barbara Sabol, also a member of the
African-American Advisory Committee supported his comments, and
said that she ``visualized a disproportionate under
representation of African-Americans. . . .'' She went on to say
that ``the Bureau placed more emphasis on reducing the cost of
the census than on reducing the differential undercount.\33\
Dr. Juliette Thorpe Okotie-Eboh, chair of the African-American
Advisory Committee, criticized the choice of counties as the
geographic level for sampling.\34\
---------------------------------------------------------------------------
\32\ Minutes of the joint meeting of the Census Advisory Committee
on African American, American Indian and Alaska Native, Asian and
Pacific Islander, and Hispanic Populations, December 11-13, 1995, p.
53.
\33\ Ibid.
\34\ Ibid.
---------------------------------------------------------------------------
At the Government Reform and Oversight hearing on February
29, 1996, following the announcement of the census 2000 plan,
Rep. Carrie Meek raised concerns about the use of sampling, and
the possibility of aggravating the undercount.\35\ Rep. Meek
subsequently introduced H.R. 3558 which would require the
Census Bureau to count 90 percent of the population at the
census tract level before sampling.
---------------------------------------------------------------------------
\35\ Transcript of testimony before the Committee on Government
Reform and Oversight February 29, 1996, pp. 54-58.
---------------------------------------------------------------------------
Others have been more strident in their criticism of
sampling. Rep. Thomas Petri said ``As many members of this
committee are aware, I believe sampling techniques should be
used only for guidance in conducting the census and not for
adjusting the final numbers. To rely on sampling rather than
the final census count would be comparable to changing election
returns if they are at variance with public opinion polls.''
\36\ Rep. Thomas Barrett said ``. . . it seems to me that
reliance on sampling, then, would allow--using your statement
of a bicoastal tendency--would allow Congress, the House of
Representatives, since you have much more electoral strength in
California, New York, Florida, those States, to use mechanisms
that we pass here in the House to benefit those States. They've
got the votes here. Let's just use a sampling that benefits
those areas of the country.'' \37\
---------------------------------------------------------------------------
\36\ Ibid., p. 31.
\37\ Ibid, pp. 42-43.
---------------------------------------------------------------------------
On the other hand, a National Academy of Sciences panel,
the General Accounting Office, and the Department of Commerce
Inspector General all endorse the use of sampling. Francis
DeGeorge, Department of Commerce Inspector General, testified
before the Government Reform and Oversight Subcommittee on
National Security, International Affairs, and Criminal Justice,
``The bureau has selected a design for the 2000 census that
includes some sampling but does not go far enough. . . .'' \38\
L. Nye Stevens, Director of Federal Management and Workforce
Issues, U.S. General Accounting Office, testified ``We are
particularly encouraged by the decision to adopt sampling among
the nonresponse population as a basic foundation of the count.
We have long advocated this step.'' \39\
---------------------------------------------------------------------------
\38\ Transcript of testimony before the Committee on Government
Reform and Oversight Subcommittee on National Security, International
Affairs, and Criminal Justice, October 25, 1995, p. 13.
\39\ L. Nye Stevens, Director of Federal Management and Workforce
Issues, U.S. General Accounting Office. Testimony before the House
Government Reform and Oversight Subcommittee on National Security,
International Affairs, and Criminal Justice, October 25, 1995.
---------------------------------------------------------------------------
In its first interim report, the National Academy of Sciences
panel on census methodology stated ``We support the use of
sampling procedures in the follow-up of households that do not
respond by mail (or telephone call) to the census. . . .'' \40\
---------------------------------------------------------------------------
\40\ White, Andrew A. And Keith F. Rust, eds., Sampling in the 2000
Census: Interim Report I, Panel to Evaluate Alternative Census
Methodologies, Committee on National Statistics, National Research
Council, National Academy Press, Washington, DC, June 1996.
---------------------------------------------------------------------------
Hon. Cardiss Collins.
Hon. Henry A. Waxman.
Hon. Tom Lantos.
Hon. Major R. Owens.
Hon. Edolphus Towns.
Hon. John M. Spratt, Jr.
Hon. Louise McIntosh Slaughter.
Hon. Gary A. Condit.
Hon. Bernard Sanders.
Hon. Karen L. Thurman.
Hon. Carolyn B. Maloney.
Hon. Barbara-Rose Collins.
Hon. Eleanor Holmes Norton.
Hon. James P. Moran.
Hon. Gene Green.
Hon. Carrie P. Meek.
Hon. Chaka Fattah.
Hon. Elijah E. Cummings.
ADDITIONAL VIEWS OF HON. THOMAS M. BARRETT
I agree with the concerns expressed in this report about
the sampling and statistical adjustments in the Census Bureau
plan for the upcoming decennial census, and the subjectivity
involved with such methodologies.
As the 1990 Census demonstrated, our current decennial
census methodology, based on actual enumeration, is far from
perfect. The 1990 Census resulted in significant undercounts in
many areas in our Nation, specifically in communities with high
numbers of minority residents.
The undercount problem must be addressed. The census
determines how entitlement and other resources will be
distributed and plays the central role in the apportionment of
Representatives in Congress. The census updates the status and
provides data about who Americans are and how Americans live.
The census forms the base core of information that affects
policy and planning decisions in all levels of our public and
private sectors. Census results affect every American. For our
Nation to obtain a useful accounting, every American resident
deserves to be counted. If we are to have a fair and just
society, the undercount problem must be resolved. Yet, we know
we will never be able to count absolutely everyone.
My concern revolves around how we remedy the undercount
problem. I strongly believe that we must mount a strong and
dedicated effort to vastly improve our system based on actual
enumeration. The way to do this is to spend the dollars
necessary for an accurate count.
In converse to actual enumeration, sampling would reduce
incentives for States and local governments to strive for an
accurate count because they could rely on the inaccuracies of
sampling to make up for deficiencies.
In these politically volatile times, I do not believe we
should go down the road of picking and choosing a census
methodology reliant on sampling and statistical adjustments
which contain subjective criteria and assumptions. A
methodology that may benefit minority groups today could be
later modified or exchanged for another that could be designed
to the detriment of specific minority populations. As we well
know, political and judicial winds do shift. The people,
however, do not go away; and their actual existence cannot be
politically modified. Therefore, we need to find ways to
actually count them.
Unlike many of my colleagues, I am not opposed to any and
all proposals to adjust the census. I also disagree with many
of my colleagues who seem satisfied with the status quo. Those
of us who oppose the Census Bureau's sampling proposal must put
the money
where our mouths are and adequately fund the Census Bureau and
State and local entities involved with census efforts. I am
disappointed that the majority of this Congress has expressed
their disagreement with the Census Bureau's sampling proposals
by slashing the Census budget.
Hon. Thomas M. Barrett.
ADDITIONAL DISSENTING VIEWS OF HON. CARRIE P. MEEK
Not fully counting African-Americans in the decennial
census originates with the Constitution. Article 1 section 2 of
the Constitution, as ratified in 1788, provides that only
three-fifths of the actual number of slaves should be counted
in the decennial census for purposes of determining the number
of Representatives each State shall have.
While this constitutionally mandated undercount of African-
Americans was repealed in 1868 by the ratification of the 14th
amendment, we continue to see its legacy in the taking of the
census. The committee's report, unfortunately, takes a
complacent view of the continuing failure to count all African-
Americans and other minorities.
In August 1992, the Census Bureau reported that the 1990
decennial census had failed to count about 4 million persons--
or about 1.6 percent of the Nation's population. This
undercount was not random. The Census Bureau further reported
that the undercount in the 1990 census was above average for
African-Americans and other minorities. For African-Americans,
the undercount was 4.4 percent. For Hispanics, the undercount
was 5 percent. For American Indians, Eskimos, and Aleuts it was
4.5 percent. For Asians and Pacific Islanders, it was 2.3
percent. The committee's report could have stated these simple
facts about the undercount, but inexplicably failed to do so.
When I was a teacher, I noticed some students sat in front
and always volunteered. Others sat in the back and never wanted
to be noticed. They were all important to me.
The Constitution now requires that we conduct the 2000
census in the same way as a good teacher conducts a class. We
must try to count everyone, not just those who are easy to
count.
The committee's report attacks the use of sampling, but the
report is unfair because it does not explain why the Census
Bureau is proposing use sampling in the 2000 census. Look at
the six findings of the report. There is no finding explaining
why the Census Bureau has recommended two types of sampling for
the 2000 census. There is no finding explaining that the Census
Bureau is proposing to use sampling for two simple reasons: to
come closer to counting everyone and to reduce costs.
One type of sampling--called Integrated Coverage
Measurement--calls for a sample of 750,000 households to
correct for the undercount that has been present in all past
censuses and was worse in the 1990 census than in the 1980
census.
This undercount deserves recognition by the committee as a
serious problem. But the 21 page single-spaced report almost
completely ignores the undercount problem.
There is no mention of the undercount on page 1.
There is no mention of it on page 2.
There is no mention of it on page 3.
There is no mention of it on page 4.
There is no mention of it on page 5.
There is no mention of it on page 6.
There is no mention of it on page 7.
There is no mention of it on page 8.
Pages 9 and 10 briefly refer to an undercount, but don't
explain what the undercount problem is.
There is no mention of it on page 11.
On page 12 there is a brief discussion of whether sampling
will help solve the undercount problem, but there is still no
explanation of what the undercount problem is.
There is no mention of it on page 13.
There is no mention of it on page 14.
There is a one-sentence reference to undercount on page 15,
but again there is no explanation of what the problem is.
There is no mention of it on page 16.
There is no mention of it on page 17.
There is no mention of it on page 18.
There is no mention of it on page 19.
Finally, in one sentence near the bottom of page 20, the
report says the committee ``is concerned that there has always
been an undercount, and that in the 1990 Census there was an
increase in the differential undercount of minorities.'' That
is the report's entire explanation of the problem that led the
Census Bureau to recommend the Integrated Coverage Measurement
type of sampling for the 2000 census.
The committee's report is backward. The problem should be
fully explained at the beginning, not barely mentioned at the
end.
Since the report almost completely ignores the undercount
problem, it is important to remember the undisputed facts about
the undercount in the 1990 census.
The Census Bureau reported in 1992 that for some States the
1990 undercount was above the national average of 1.6 percent.
For example, in California about 835,000 people--or 2.7 percent
of the population--were not counted. In New Mexico about 48,000
people--or about 3.1 percent of the population--were not
counted. In Florida about 260,000 people--or about 2 percent of
the population--were not counted.
The Census Bureau also reported in 1992 that the 1990
undercount for certain areas within a particular State was well
above the national average of 1.6 percent. For example, in Long
Beach City, CA, 17,000 people--or 3.7 percent of the
population--were not counted. In Dade County, FL, 74,000
people--or 3.7 percent of the population--were not counted. In
Fairfax County, VA, 15,000 people--or 1.8 percent of the
population--were not counted.
The other type of sampling proposed by the Census Bureau is
the so-called ``sampling for nonresponse.'' For the 2000 census
the Bureau proposes taking a 1 in 10 sample of nonrespondents
after 90 percent of the households in a county have responded.
The Census Bureau estimates that this type of sampling will
save $500 million as compared to trying to count everyone
directly.
I have criticized the details of this proposal, and my
bipartisan bill, H.R. 3558, would require that the sampling be
done within census tracts and not within counties.
But criticizing the details of a sampling proposal is a far
cry from a wholesale condemnation of any type of sampling.
The committee's report does not recognize that many experts
support the use of sampling in the 2000 census if it is done
carefully. The June 1996 Interim Report of the Committee on
National Statistics of the National Research Council, entitled
Sampling in the 2000 Census, concluded that ``A combination of
sampling for nonresponse follow-up and for integrated coverage
measurement is key to conducting a decennial census at an
acceptable cost, with increased accuracy and overall quality,
and reduced differential undercoverage.'' The September 1996
report of a panel of the American Statistical Association
concludes that the use of sampling in the 2000 census ``has the
potential to increase the quality and accuracy of the count and
reduce costs.''
The committee's report is opposed by many groups whose
members will use the 2000 Census. A letter form Ann Azari, the
mayor of the city of Fort Collins, states that the Advisory
Committee opposes the committee's report and supports the use
of sampling to improve the accuracy and reduce the cost of the
2000 census. The members of the Advisory Committee include the
American-Arab Anti-Discrimination Committee, the Business
Roundtable, the National Association for the Advancement of
Colored People, the National Association of Counties, the
National Association of Towns and Townships, the National
Conference of State Legislatures, the National Council of
LaRaza, the National Congress of American Indians, the National
Governors Association, the National League of Cities, the U.S.
Chamber of Commerce, and the U.S. Conference of Mayors. The
Japanese American Citizens League, in a separate letter, says
it opposes the committee's report because barring the use of
sampling in the 2000 census means that ``the outcome of the
2000 census will be no different from those past: unequal and
unfair.''
In conclusion, the committee's report is a one-sided attack
on any use of sampling in the 2000 census. The report does not
fairly reflect the concerns of those members of the committee
who believe that the careful use of sampling in the 2000 census
will help this Nation erase the bitter legacy of the original
constitutional mandate to not fully count African-Americans.
Hon. Carrie P. Meek.
-