Data Quality: Improvements to Count Correction Efforts Could	 
Produce More Accurate Census Data (20-JUN-05, GAO-05-463).	 
                                                                 
The U.S. Census Bureau (Bureau) conducted the Count Question	 
Resolution (CQR) program to correct errors in the count of	 
housing units as well as dormitories and other group living	 
facilities known as group quarters. GAO was asked to assess	 
whether CQR was consistently implemented across the country,	 
paying particular attention to whether the Bureau identified	 
census errors that could have been caused by more systemic	 
problems. GAO also evaluated how well the Bureau transitioned to 
CQR from an earlier quality assurance program called Full Count  
Review. 							 
-------------------------Indexing Terms------------------------- 
REPORTNUM:   GAO-05-463 					        
    ACCNO:   A27057						        
  TITLE:     Data Quality: Improvements to Count Correction Efforts   
Could Produce More Accurate Census Data 			 
     DATE:   06/20/2005 
  SUBJECT:   Census						 
	     Data collection					 
	     Data integrity					 
	     Errors						 
	     Internal controls					 
	     Population statistics				 
	     Program evaluation 				 
	     Program management 				 
	     Quality assurance					 
	     Statistical data					 
	     2000 Decennial Census				 
	     2010 Decennial Census				 
	     Census Bureau Count Question Resolution		 
	     Program						 
                                                                 
	     Census Bureau Demographic Full Count		 
	     Review						 
                                                                 

******************************************************************
** This file contains an ASCII representation of the text of a  **
** GAO Product.                                                 **
**                                                              **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced.  Tables are included, but    **
** may not resemble those in the printed version.               **
**                                                              **
** Please see the PDF (Portable Document Format) file, when     **
** available, for a complete electronic file of the printed     **
** document's contents.                                         **
**                                                              **
******************************************************************
GAO-05-463

                 United States Government Accountability Office

                     GAO Report to Congressional Requesters

June 2005

DATA QUALITY

Improvements to Count Correction Efforts Could Produce More Accurate Census Data

                                       a

GAO-05-463

[IMG]

June 2005

DATA QUALITY

Improvements to Count Correction Efforts Could Produce More Accurate Census Data

  What GAO Found

The CQR program, which ran from June 30, 2001, to September 30, 2003,
played an important role in improving the quality of data from the 2000
Census in that it corrected numbers affecting 47 states and over 1,180
governmental units. Although this is a small percentage of the nation's
more than 39,000 government entities, the count revisions impacted private
homes, prisons, and other dwellings and, in some cases, were significant.
For example, when the Bureau deleted duplicate data on students at the
University of North Carolina at Chapel Hill and made other corrections,
that state's head count dropped by 2,828 people. Similarly, CQR found that
more than 1,600 people in Morehead, Kentucky, were counted in the wrong
location.

GAO identified several shortcomings with the CQR program, including
inconsistent implementation by the Bureau's regional offices and the
posting of inaccurate data to the Bureau's Web-based errata report.
Moreover, while CQR found the counting of group quarters to be
particularly problematic, the Bureau did not perform an active, nationwide
review of these known trouble spots, and thus missed an opportunity to
potentially improve the accuracy of the data for these dwellings. Further,
because CQR had more stringent documentation requirements compared to a
preceding program called Full Count Review, CQR rejected hundreds of
unresolved full count issues, missing another opportunity to improve the
data. As its plans proceed for the 2010 Census, it will be important for
the Bureau to address the operational issues GAO identified. Moreover,
because the data for apportionment and redistricting were later found to
be flawed for some jurisdictions, it will be important for the Bureau to
develop a count correction program that is designed to systematically
review and correct these essential figures.

United States Government Accountability Office

Contents

  Letter

Results in Brief
Background
Scope and Methodology
CQR Program Corrected Numerous Data Errors, but More

Consistent Implementation and Other Improvements Are Needed Better
Strategic Planning and Other Actions Could Improve Future

Count Correction Efforts Conclusions Recommendations for Executive Action
Agency Comments and Our Evaluation

1 2 4 6

8

21 26 27 28

Appendixes

Appendix I:

Appendix II:

Appendix III: Appendix IV:

Change in State Populations As a Result of Count Question Resolution
Program

Human Error and Other Factors Contributed to University of North Carolina
Counting Errors

Comments from the Department of Commerce

GAO Contact and Staff Acknowledgments

31

33 35 40

Figures Figure 1:

Figure 2: Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Time Line Showing Relationship of CQR Program to Key
Census 2000 Milestones 5
Map of Census Bureau's 12 Regions 7
CQR Revisions Affected Numerous Governmental Units
in Most States 10
Students in 26 UNC Dormitories Were Counted Twice in
the Census 12
Prisoners in Cameron, Missouri, Were Mistakenly
Omitted From the Town's Population Count 13
CQR Table in the Bureau's 2000 Notes and Errata Report
Showing Faulty Links to Data 18
Initial Census Data on Bureau Web Site Do Not Inform
Users That Some Numbers Have Been Revised 20

Contents

This is a work of the U.S. government and is not subject to copyright
protection in the United States. It may be reproduced and distributed in
its entirety without further permission from GAO. However, because this
work may contain copyrighted images or other material, permission from the
copyright holder may be necessary if you wish to reproduce this material
separately.

A

United States Government Accountability Office Washington, D.C. 20548

June 20, 2005

The Honorable Wm. Lacy Clay
Ranking Minority Member
Subcommittee on Federalism and the Census
Committee on Government Reform
House of Representatives

The Honorable Carolyn B. Maloney
House of Representatives

Complete and accurate data from the decennial census are central to our
democratic system of government. As required by the Constitution, census
results are used to apportion seats in the House of Representatives.
Census
data are also used to redraw congressional districts, allocate billions of
dollars in federal assistance to state and local governments, and for many
other public and private sector purposes. Failure to deliver quality data
could skew the equitable distribution of political power in our society,
impair public and private decision making, and erode public confidence in
the U.S. Census Bureau (Bureau).

To ensure it delivers accurate data, the Bureau employs a number of
quality
assurance programs throughout the course of the census. One such effort
during the 2000 Census was the Count Question Resolution (CQR)
program, which enabled state, local, and tribal governments to formally
challenge the counts of housing units and "group quarters" (dormitories,
prisons, and other group living facilities), and their associated
populations.
Bureau personnel could initiate a review of the counts as well.

Although the Bureau did not design CQR with the intention of
incorporating any of the corrections that resulted from it into Census
2000
data products-including the numbers used for congressional
apportionment and redistricting (figures commonly referred to as "public
law data")-governmental entities could use the updated information when
applying for federal aid that uses census data as part of an allocation
formula, as well as for other purposes. Because the count corrections
could
have political and financial implications for states and localities, it
was
important for the Bureau to carry out CQR consistent with its protocols.
CQR began on June 30, 2001, and no new submissions were accepted after
September 30, 2003.

This letter responds to your request to review the conduct of the CQR
program. As agreed with your offices, we reviewed the results of the CQR
program and assessed whether the program was consistently implemented
across the country. In doing this, we paid particular attention to the
extent to which the Bureau reviewed the census data for errors that could
have been caused by broader, more systemic problems. We also evaluated how
well the Bureau transitioned to CQR from an earlier quality assurance
program called Full Count Review.

To meet these objectives, we reviewed relevant program documents and
examined case files and conducted on-site inspections at four of the
Bureau's regional offices where some of the largest CQR corrections took
place. We also interviewed officials and staff responsible for
administering the CQR program at the Bureau's headquarters and 12 regional
offices. We did our audit work between February 2004 and March 2005 in
accordance with generally accepted government auditing standards.

Results in Brief	The CQR program corrected data affecting over 1,180 of
the nation's more than 39,000 governmental units including states,
counties, and cities. Although the national and state-level revisions were
relatively small, in some cases the corrections at the local level were
substantial. For example, CQR increased Morehead, Kentucky's, population
total by more than 1,600 people because the Bureau mistakenly attributed
local university students, who lived in dormitories located within the
city, to the population count of an unincorporated section of the county
in which Morehead is located. Likewise, the Bureau added almost 1,500
persons to the population count of Cameron, Missouri, when CQR found that
a prison's population was erroneously omitted.

That said, we also found critical aspects of the CQR program in need of
improvement. For example, CQR was not consistently implemented by the
Bureau's regional offices. Only the Bureau's Los Angeles Regional Office
appeared to do any comprehensive, systematic research to identify possible
count errors beyond those that were submitted by governmental units. Had
it not been for Los Angeles' self-initiated review, several data
errors-including instances where college dormitories were counted in the
wrong geographic location-would have remained uncorrected because they
were not identified by the affected jurisdiction.

In contrast, the Bureau's Charlotte Regional Office found that almost
2,700 students were counted twice at the University of North Carolina,

the discovery of which came about largely because two key census employees
in Charlotte were alumni of the school and curious to see whether
dormitories there were enumerated correctly. One factor behind the
disparate execution of the CQR program seems to have been vague and
sometimes inconsistent guidance and training that left staff in the
regional offices with different understandings of whether they could
conduct selfinitiated research.

In addition, although the Bureau maintained an errata report on its Web
site that listed the CQR revisions to the census data, our partial review
of that information found several discrepancies between the updated
figures, and what the numbers should have been. For example, the revised
number of housing units for Sioux Falls, South Dakota, was almost 47,000
units too low. Likewise, the errata data on the total housing unit count
for Burlington County, New Jersey, mistakenly excluded about 145,000
units. Moreover, embedded links on the Web site that were supposed to take
users to revisions at lower levels of geography did not always work and
produced error messages instead.

The CQR program was also poorly integrated with its predecessor program,
Full Count Review. Although the Bureau planned to fold unresolved full
count issues into CQR, the latter program had more rigorous documentation
requirements. Consequently, hundreds of unresolved Full Count Review cases
lacked CQR's necessary documentation, were rejected from CQR, and received
no further review.

Overall, the CQR program was an important quality assurance tool, but the
Bureau needs to address the operational issues we identified. Further,
given the growing challenges to counting the nation's population, census
errors are inevitable, and as the Bureau makes plans for the 2010 Census,
it will be important for it to have a mechanism specifically designed to
methodically review and correct errors in the public law data and
subsequent data releases to the greatest extent possible. The lessons the
Bureau has learned from CQR should provide valuable experience in
developing such a program.

With that in mind, we recommend that the Secretary of Commerce direct the
Bureau to improve its count correction efforts by taking such actions as:
(1) consolidating Full Count Review and CQR into a single program that
systematically reviews and corrects any errors prior to the release of the
public law data; (2) expediting count correction efforts, in part, by
using enumerators to help investigate data discrepancies while conducting
their

field work; (3) prioritizing the investigation of data challenges based on
the magnitude of the suspected error; (4) ensuring the accuracy and
accessibility of the revised data on its Web site; and (5) improving
training and guidance provided to regional offices to help ensure count
correction activities are consistently implemented.

The Acting Deputy Secretary of Commerce provided written comments on a
draft of this report (see app. III). Commerce acknowledged that "the
report provides a good overview of program results and makes several
useful observations and recommendations," and agreed with our finding that
the process for conducting internal reviews was not consistently
implemented. Nevertheless, Commerce took exception to our recommendations
calling for the Bureau to design a count correction effort capable of
identifying and correcting errors in the apportionment and redistricting
data before that critical information is released. Commerce noted that
such an approach was infeasible largely because of time and logistical
constraints. Our report recognizes these challenges; further, the steps we
recommend could help the Bureau overcome these very challenges and deliver
more accurate public law data.

Background	The Bureau launched the CQR program on June 30, 2001, as the
last in a series of quality assurance initiatives aimed at improving the
accuracy of 2000 Census data (see fig. 1). Specifically, the CQR program
provided a mechanism for state, local, and tribal governments to have the
Bureau correct errors in certain types of census data. The Bureau referred
to these challenges as "external cases." Bureau personnel could also
initiate reviews of suspected count errors, independent of these
challenges, for further review. These were known as "internal cases." Many
of the internal cases were unresolved issues inherited from Full Count
Review. Indeed, when the Full Count Review program began, the Bureau
planned to fold unresolved issues from that program into CQR. The Bureau
accepted no new submissions after the program officially ended on
September 30, 2003, although it continued to review challenges submitted
before the deadline and completed the final revisions in the summer of
2004.

Figure 1: Time Line Showing Relationship of CQR Program to Key Census 2000
Milestones

A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A
M J J A S

         Census Day-             Full Count                       CQR program 
        April 1, 2000                                       officially ended- 
                                   began-                       September 30, 
                                  June 2000                              2003 

Nonresponse follow-up began- April 2000

                                  CQR program

Apportionment data sent to the President- Dec. 28, 2000 began- June 30,
2001

Source: GAO analysis of U.S. Census Bureau data.

Three types of corrections were permissible under the CQR program: (1)
boundary corrections, where a jurisdictional boundary of a functioning
governmental unit was in the wrong location; (2) geocoding corrections,
where the Bureau placed the living quarters and their associated
population in the wrong location; and (3) coverage corrections, where the
Bureau properly enumerated specific living quarters and their
corresponding population during the census but incorrectly added or
deleted the information during data processing. Bureau officials were to
research cases using existing Bureau data gathered during the 2000 Census;
they could not conduct any new fieldwork to resolve count questions. The
Bureau required governmental entities to accompany their challenges with
specific documentation before it would investigate their claims.

Importantly, under the design of CQR, if a governmental unit had evidence
that the Bureau missed housing units or group quarters that existed on
Census Day 2000 (April 1), but the Bureau's records indicated that all of
the Bureau's boundary information, geocoding, and processing were properly
implemented, the Bureau would not change the data. Rather, the Bureau was
to address this as part of the planning process for the 2010 Census.

If the CQR program corrected the population or housing unit counts of a
particular entity, the Bureau was to issue revised, official figures for
that

jurisdiction. The governmental unit could then use the updated numbers for
future programs requiring 2000 Census data. CQR corrections were also used
to modify annual post-censal estimates beginning December 2002 and were
publicized on the Bureau's Census 2000 and American FactFinder Web sites
(www.census.gov and www.factfinder.census.gov, respectively), as part of
the 2000 Census notes and errata. However, CQR was not designed or
publicized as a mechanism to correct the census results for purposes of
apportionment and redistricting.

In compliance with legal requirements, the Bureau produced apportionment
data by December 31, 2000, and redistricting data by April 1, 20011 (this
information is known collectively as public law data). Although the law
does not require that states use census data to redraw the boundaries of
congressional districts, most states have always done so. Nothing would
preclude the states from using the corrected data for redistricting. The
general perception of the impartiality of the Bureau and the great cost
and administrative effort required to take a census have been strong
arguments in favor of using the Bureau's data.

Scope and Methodology

As agreed with your offices, we assessed whether the program was
consistently implemented across the country, paying particular attention
to the extent to which the Bureau reviewed the census data for errors that
could have been caused by broader, more systemic problems, such as
shortcomings with a particular census-taking procedure. We also evaluated
how well the Bureau transitioned from an earlier quality assurance program
used in the 2000 Census, Full Count Review.

To assess the implementation of the CQR program, we obtained a
headquarters perspective by reviewing program documents and case files at
the Bureau's offices in Suitland, Maryland, as well as program results
reported on the Bureau's Web site.2 As part of this assessment, we
reviewed the program's internal controls, especially those controls
related to ensuring data quality. We also interviewed Bureau officials
responsible for administering the program.

113 U.S.C. S:S: 141(b) - (c).

2Based on our assessment of the data, we found the case file information
and program results sufficiently reliable for our review.

To determine how CQR was implemented in the field, we visited 4 of the
Bureau's 12 regional offices-Charlotte, North Carolina; Denver, Colorado;
Kansas City, Missouri; and Los Angeles, California (see fig. 2).

Figure 2: Map of Census Bureau's 12 Regions

Source: GAO analysis of U.S. Census Bureau data.

We selected these regions because they included the six states and 10
governmental units within those states where the largest CQR count
revisions occurred. We supplemented these cases by selecting an

additional seven states and 61 places within the four regions for further
examination. The 61 localities were selected because they represented the
full spectrum of CQR cases and were geographically diverse.

At each of the four regions, we reviewed regional case file information
and interviewed Bureau personnel responsible for implementing CQR, such as
program managers and geographers. We also made a site visit to at least
one type of facility-including prisons, apartment buildings, and
dormitories-in each region to understand firsthand the nature of the
errors and the corrections made. To augment these four regional visits and
obtain a more complete picture of how CQR was implemented, we used a
structured telephone interview to elicit information from program
officials at the Bureau's eight remaining regional offices that we did not
visit in person.

To determine the extent to which the Bureau reviewed census data for
systemic errors, and its procedures for folding unresolved cases from the
Full Count Review program into CQR, we examined program manuals,
memoranda, and other documents, and interviewed officials in the Bureau's
headquarters and all of its regional offices. As part of this effort, we
also analyzed the CQR case-tracking data in an attempt to determine the
number of unresolved Full Count Review cases that were rolled into the CQR
program. However, we were unable to do this because the tracking system
did not contain information on which CQR cases originated as full count
issues.

We requested comments on a draft of this report from the Secretary of
Commerce. On May 20, 2005, we received the Acting Deputy Secretary's
written comments and have reprinted them in appendix III; we address them
in the Agency Comments and Our Evaluation section of this report.

CQR Program Corrected Numerous Data Errors, but More Consistent
Implementation and Other Improvements Are Needed

Overall, the CQR program corrected data affecting over 1,180 of the
nation's more than 39,000 governmental units. The revisions impacted a
range of housing types including private homes with only a handful of
residents, to college dormitories and prison cell blocks with populations
in the thousands. At the same time, however, we identified several
shortcomings with the CQR program, including inconsistent handling of
internal cases by the Bureau's regional offices and inaccurate data being
posted to the Bureau's public Web site. Moreover, while CQR found the
counting of group quarters in their correct location-a problem known as
geocoding error-to be particularly challenging, the Bureau did not

perform a nationwide review of these known trouble spots, and thus missed
an opportunity to improve the accuracy of the data for these dwellings.

CQR Program Corrected Nationwide, the CQR program corrected count errors
involving Errors in Hundreds of governmental units in 47 states, Puerto
Rico, and the District of Columbia Governmental Units (see fig. 3).3 Three
states-Maine, New Hampshire, and Rhode Island-had

no CQR corrections.

3The Bureau also made corrections to governmental units classified as
American Indian/Alaska Native Areas.

Figure 3: CQR Revisions Affected Numerous Governmental Units in Most
States

Source: GAO analysis of U.S. Census Bureau data.

aThe Bureau made count changes in Hawaii and the District of Columbia, but
these revisions were made at the census block level and did not change the
state's governmental unit counts.

The corrections affected over 1,180 governmental units in the United
States. Although this is a small percentage of the nation's more than
39,000 governmental units, the impact of those changes on local
governments was, in some cases, substantial, and could have implications
for federal assistance and state funding programs that use census numbers
in their allocation formulas, as well as other applications of census
data.

For example, officials in one Kentucky county challenged the geocoding of
a housing unit located near new precinct and congressional district
boundaries. They told the Bureau that the new boundaries split the county,
and they were concerned that the geocoding error would affect where the
housing unit's few occupants registered to vote. Because the housing unit
was improperly geocoded, the Bureau corrected the data.

With respect to fiscal effects, the Controller of the State of California
uses population figures as the basis for refunding a portion of state
taxpayer fees-including automobile licensing fees-to cities and counties.
Because of an error in the 2000 Census, Soledad, California, officials
estimated it lost more than $140,000 in state refunds when over 11,000
residents were incorrectly counted in two nearby cities' populations,
according to city and state officials. Although CQR eventually corrected
the error, Soledad did not recover the funds that went to the other
cities.

Other examples of large CQR corrections include the following (See app. I
for a complete list of state-level population changes):4

o 	North Carolina's population count was reduced by 2,828 people, largely
because the Bureau had to delete duplicate data on almost 2,700 students
in 26 dormitories (see fig. 4) at the University of North Carolina (UNC)
at Chapel Hill. The erroneous enumerations occurred, in large part,
because of mistakes that occurred in various preparatory activities
leading up to the 2000 Census. (See app. II for a more detailed discussion
of this incident.)

4Because a population increase in one government entity was typically
offset by a loss in population in a neighboring entity (or vice-versa),
there was generally little net change in population counts at the national
and state levels as a result of the CQR program and no effect on
apportionment.

Figure 4: Students in 26 UNC Dormitories Were Counted Twice in the Census

Source: GAO.

o 	The population count of Morehead, Kentucky, increased by more than
1,600 when CQR found that a large number of students from Morehead State
University's dormitories were erroneously excluded from the city's
population. During the 2000 Census, the Bureau had incorrectly identified
the dormitories as being outside city limits and in an unincorporated area
of Rowan County.

o 	The population count of Cameron, Missouri, was off by nearly 1,500
people when the Bureau found that the prison population of the state's
Crossroads Correctional Center was inadvertently omitted from the town's
headcount (see fig. 5). The correction to the town's population accounted
for the entire 1,472 person increase in Missouri's total population under
the CQR program.

Figure 5: Prisoners in Cameron, Missouri, Were Mistakenly Omitted From the
Town's Population Count

                                  Source: GAO.

o 	The population of the city of Waseca, Minnesota, increased by more than
1,100. The 2000 Census had mistakenly included the prison population for
the Waseca Federal Correctional Institute in two surrounding townships in
Waseca County. The CQR program resulted in the population being shifted to
the city.

o 	The population of Colorado increased by more than 750 in large part
because a processing error in counting housing units in Grand Junction
initially excluded almost 700 people from the city's population total.

o 	The population of Denver and Arapahoe Counties in Colorado shifted by
more than 900 because the Bureau had incorrectly assigned the location of
two apartment complexes. As a result of CQR, the apartment complexes were
incorrectly identified and counted as being in Denver but under CQR were
later found to be in adjoining Arapahoe County.

CQR Program Was Unevenly The Bureau's 12 regional offices did not always
adhere to the same set of

Executed 	procedures when developing internal cases, and this, in turn,
produced uneven results. Importantly, the procedures used to execute
public programs need to be well documented, transparent, and consistently
applied in order to ensure fairness, accountability, defensible decisions,
and reliable outcomes. To do otherwise could raise equity questions.

One variation in the way internal cases were handled was evident at the
Bureau's Los Angeles Regional Office, which appeared to be the only region
to do comprehensive methodical research to actively identify possible
count errors beyond those that were submitted by governmental entities.
According to the office's senior geographer, the geography staff developed
a structured approach to systematically examine census data from all the
prisons and colleges within the office's jurisdiction, because the data on
both types of group quarters were known to be problematic. He added that
the more problems they found, the more they were motivated to keep
digging. The geographer noted that the in-depth review was possible
because the Los Angeles region covers only the southern half of California
and the state of Hawaii, and thus has fewer governmental units compared to
the Bureau's other regional offices.

Had it not been for the Los Angeles region's self-initiated and systematic
review, certain data errors would have gone uncorrected because they were
not identified by the affected jurisdiction. For example, regional staff
found instances where college dormitories were counted in the wrong
geographic location, which, in turn, affected the population counts of
their surrounding locales. Such was the case with California State
University Monterey Bay (CSUMB) and the University of California at Santa
Barbara (UCSB). As a result, the Bureau transferred a population of more
than 1,400 between the towns in which they were initially counted and in
which CSUMB is located and shifted a population of more than 2,700 between
the city and the unincorporated area of the county in which UCSB is
located.

The Bureau's Charlotte Region, while also more active than the Bureau's
other offices that generated internal CQR cases, seemed to be less
methodical and comprehensive than Los Angeles in its approach. For
example, although Charlotte geographers detected the duplicate count of
almost 2,700 students at the University of North Carolina mentioned in the
previous section, their research was not the result of any systematic
review. Rather, it came about largely because of the curiosity of key
employees in the Charlotte office, who were also alumni of the school.
(See app. II for more details on the circumstances surrounding the
duplicate count.)

Better Guidance and Vague guidance was one reason for the disparate
handling of internal cases. Training Could Improve For example, the CQR
procedural manual indicates that the Bureau's Implementation 12 regional
offices were to research CQR cases "as appropriate." However,

the manual did not define whether this meant that the regional offices

should initiate their own data reviews or merely verify CQR cases
submitted by governmental units. More generally, numerous geographers we
interviewed-the primary users of the manual-did not find it userfriendly,
noting it was confusing, complex, or impractical. For example, a
geographer pointed out that the manual did not have an index covering the
eight chapters and 26 appendixes, which would have helped them more
quickly find information and procedures. In addition, we found that the
manual and other documents did not discuss how program staff were to
address Full Count Review issues.

The Bureau's training was also problematic and likely added to the
implementation disparities. For example, geographers in five regions told
us that during training they were instructed or given the impression they
were not to generate additional internal cases beyond the small number of
count errors that had already been identified at the beginning of the
program. Also, geographers in two of these regions told us they were
specifically told not to investigate any count errors they found that were
outside the scope of the cases that governmental units submitted.
Conversely, geographers in the other seven regions said they were not
restricted in any way.

There were other training issues as well. According to the Bureau's draft
CQR program assessment, the final version of which is pending, some
training materials were developed at the last minute and were never
finalized, and training began before needed software was in place at all
the research divisions. Proper training was particularly important
because, as the draft evaluation notes, staff assigned to the CQR program
had census experience but limited geographic and field operations
knowledge. Others had limited or no Census 2000 software program
experience.

Internal Control and Quality Federal internal control standards call on
agencies to employ edit checks Assurance Problems Led the and other
procedures to ensure their information processing systems Bureau to Report
Erroneous produce accurate and complete data.5 However, the Bureau's
internal

controls in reporting CQR results were insufficient in that we found,
after a

Data	partial review, a number of instances where the Bureau disseminated
inaccurate data on its Web site where it maintains an errata report that
lists the CQR revisions to the 2000 Census data.

5GAO, Internal Control Management and Evaluation Tool, GAO-01-1008G
(Washington, D.C.: Aug. 1, 2001).

Specifically, after comparing data from the errata report to the certified
numbers in the CQR case files, we found errors with the reporting of CQR
housing, group quarters, and population counts. Importantly, our review
found that the revised, certified figures the Bureau provided to affected
jurisdictions were correct. This is significant because the affected
jurisdictions could use these updated numbers for revenue sharing and
other programs that require census data. However, users who obtain
information from the Bureau's errata report-these can be people in
academia, government, and the private sector-would not have the most
up-to-date information. For example:

o 	The original state-level total housing unit count for Delaware
mistakenly excluded 30,000 housing units.

o 	The revised housing unit count for the Minnehaha County portion of
Sioux Falls, South Dakota, was underreported by almost 47,000 housing
units.

o 	The Burlington County, New Jersey, revised total housing count
mistakenly excluded about 145,000 units.

o 	The errata report excluded 8 of the 12 American Indian and Alaska
Native Areas that had revisions to their housing, group quarter, or total
population counts.

The Bureau later corrected these errors after we brought them to its
attention.

Although the Bureau had controls in place to ensure accurate research and
reporting, the problems we found point to the need for the Bureau to
tighten its procedures to prevent mistakes from slipping into its data
products. For example, the CQR manual included some quality control steps,
such as having headquarters divisions review field research and results.
Further, field geographers told us they consulted one another about
questions or procedures and checked each other's work, and Bureau program
managers had procedures in place to review final revisions and certify
them as correct.

Documents in the CQR case files we reviewed substantiate these practices.
Also, managers told us they randomly checked data entered into the files
that are the basis for the revisions posted to the errata report. Still,
the number of errors we found after only a partial review of the errata
files

raises questions about how effectively the Bureau implemented the quality
assurance procedures, as well as the quality of the data we did not
review. It also underscores the importance of adequate control activities
to prevent these problems from recurring.

Web-based Errata Report Should Be More User-Friendly

Data users may have encountered problems trying to access certain
information from the Web-based errata report. Additionally, because there
was no link or cross-walk between some of the initial population data the
Bureau released and the CQR revisions, users may have been unaware that
some of the original numbers had been revised.

As shown in figure 6, the Bureau's Web-based errata report presented
revised data for states, American Indian and Alaska Native Areas, and
other jurisdictions at the state or similar geographic level. Although the
table had embedded links that were supposed to take users to revisions at
lower levels of geography, these links did not always work and produced
error messages instead. We found that unless the users' software and
Internet access paralleled the Bureau's, users could not access the more
detailed data using the embedded links. Bureau staff involved with posting
the data to the Web site stated in the summer of 2004 that they were aware
of the problem, but as late as March 2005, the problem had yet to be
fixed.

Figure 6: CQR Table in the Bureau's 2000 Notes and Errata Report Showing Faulty
                                 Links to Data

            Source: GAO analysis of U.S. Census Bureau information.

At the same time, the CQR revisions may not be evident to users who access
certain data from the 2000 Census data posted elsewhere on the Bureau's
Web site. This is because these sites lack notes or flags informing users
that updated figures are available in the census errata report.6 For
example, the Bureau's American FactFinder Web site-the Bureau's primary
electronic source of census data-does not inform users that revised data
on group quarter counts, including the number of correctional
institutions, as well as data on their associated populations, exist as
part of the Bureau's notes and errata report.

American FactFinder presents data known as Summary File 1 (SF-1), which is
the first data set the Bureau produces from the census, and is used for
purposes of apportionment and redistricting. While the SF-1 data remain
unchanged, other data users may find the revised numbers better suited to
their needs. Figure 7 illustrates the existence of two sets of numbers
without any explanation. Summary File data from American FactFinder show
the population for Soledad, California, as 11,263. However, the Bureau's
errata report, which reflects the CQR revisions, shows the Soledad
population at 23,015. Because American FactFinder lacks notes or links
that tell users about the revised data, users might inadvertently obtain
erroneous information.

6The Bureau and American FactFinder home pages do not list or provide
direct links to the 2000 Census notes and errata report or the CQR program
Web site. However, the Bureau's Census 2000 Gateway Web site provides
links to both of them and its American FactFinder Web site provides an
indirect link to the notes and errata through that site's data sets.

 Figure 7: Initial Census Data on Bureau Web Site Do Not Inform Users That Some
                           Numbers Have Been Revised

Source: GAO analysis of U.S. Census Bureau information.

According to Bureau officials, while they thought about adding notes
directing users to the CQR revisions, they decided against it because they
thought it would confuse more people than it would help. They reasoned
that knowledgeable users, such as county planners and state data center
staff, are likely aware of the CQR information and would therefore not
need to be informed about the existence of the notes and errata Web site.

CQR Errors Highlight Problems with 2000 Census Address List Development
Procedures

The errors uncovered by the CQR program highlight some of the limitations
in the way in which the Bureau builds its address list for the decennial
census, particularly in the procedures used to identify and locate group
quarters. For the 2000 Census, the Bureau had three operations that were
primarily designed to locate these types of dwellings. However, given the
number of prisons and other group quarters geocoded to the wrong location,
refinements are needed. Moreover, the Bureau's draft CQR program
assessment found that the Bureau's Master Address File had numerous data
entry errors including incorrect spellings, geocoding, and zip codes.

To its credit, the Bureau is planning several improvements for 2010,
including integrating its housing unit and group quarter address lists.
This could help prevent the type of duplicate counting that occurred at
UNC when the same dormitories appeared on both lists. Likewise, the
Bureau's planned use of a satellite-based navigational system could help
census workers more precisely locate street addresses.

Better Strategic The CQR program was preceded by a quality assurance
program called Full

Count Review, which ran from June 2000 through March 2001 and,
likePlanning and Other CQR, was designed to find problems with the census
data. However, Actions Could Improve although the Bureau planned to fold
unresolved full count issues into CQR, Future Count many full count issues
were rejected from CQR because the latter program

had more stringent documentation requirements. As a result, the
BureauCorrection Efforts was unable to resolve hundreds of additional data
issues.

Numerous Unresolved Full Under the Full Count Review program, analysts
were to identify data Count Issues Could Not Be discrepancies to clear
census data files and products for subsequent Folded into CQR as Planned
processing or public release. Analysts did so by checking the data for
their

overall reasonableness, as well as for their consistency with historical
and

demographic census data and other census data products. The types of

issues flagged during Full Count Review included potential discrepancies
involving the counts and/or locations of group quarters, housing units,
and individual households, among others.

As we noted in our July 2002 report, Full Count Review identified 4,809
potential data anomalies.7 However, of these, just five were corrected
prior to the December 31, 2000, release of apportionment data and the
April 1, 2001, release of redistricting data. The corrections included a
military base, a federal medical center, and multiple facilities at two
prisons and a college that were counted in the wrong locations. That the
public law data were released with numerous data issues of unknown
validity, magnitude, and impact, gave us cause for concern, and we noted
that the Bureau missed an opportunity to verify and possibly improve the
quality of the information.

When the Full Count Review program began, the Bureau planned to fold
unresolved issues from that program into CQR. Indeed, according to a June
2000 memo on CQR policy agreements, "a by-product of [Full Count Review]
is documentation of unresolved issues for potential use in CQR." However,
because the CQR program had more rigorous documentation requirements
before it would accept a case compared to Full Count Review, a number of
issues that were deemed suitable for Full Count Review but were
unresolved, were rejected from CQR.

Of the 4,804 issues remaining after Full Count Review, 2,810 issues (58
percent), were not referred to CQR. Of the 1,994 issues (42 percent) that
were referred to CQR, 537 were actually accepted by the program. The
remaining 1,457 issues referred to CQR did not meet the Bureau's CQR
documentation requirements and, consequently, the Bureau took no further
action on them.

The Full Count training materials we examined as part of our 2002 review
did not provide any specific guidance on the type of evidence analysts
needed to support data issues. Rather, the materials instructed analysts
to supply as much supporting information as necessary. In contrast, the
CQR program had more rigorous documentation requirements. Guidance
available on the Bureau's Web site required governmental units to supply
maps and other evidence specific to the type of correction they were
requesting, or the Bureau would not investigate their submissions.

7GAO, 2000 Census: Refinements to Full Count Review Program Could Improve
Future Data Quality, GAO-02-562 (Washington, D.C.: July 3, 2002).

Simply put, Full Count Review identified hundreds of data issues but
lacked the time to investigate the vast majority of them. Then, when the
remaining cases were referred to CQR, most were rejected because they
could not meet CQR's higher evidentiary bar.

A Mechanism for Correcting Public Law Data Will Be Critical for Future
Enumerations

The Bureau lacked a program specifically designed to correct individual
count errors contained in the apportionment and redistricting data.
Because these numbers were later found to be flawed for some
jurisdictions, as the Bureau proceeds with its plans for the 2010 Census,
it will be important for it to explore options for reviewing and
correcting this essential information before it is released.

Precision is critical because, in some cases, small differences in
population totals could potentially impact apportionment and/or
redistricting decisions. For example, according to an analysis by the
Congressional Research Service, under the formula used to apportion seats
in the U.S. House of Representatives, had Utah's population count been 855
persons higher, it would have gained an additional congressional seat and
North Carolina would have lost a seat. However, had the duplicate UNC
count and other errors detected by the CQR program as of September 30,
2003, been uncovered prior to the release of the public law data, the
already narrow margin determining whether Utah gained a House seat would
have dropped to 87 persons.8 Although in this particular instance there
would not have been a change in congressional apportionment, it
illustrates how the allocation of House seats can be determined by small
differences in population counts.

Other Aspects of CQR Could Have Been Better Planned

Better planning could have improved the CQR program in other ways. For
example, the Bureau's draft evaluation of CQR found, among other issues,
that the three teams working on the planning and development phases of CQR
should have tested implementation plans earlier in the process, and
training materials were not based on the Bureau's experience in conducting
the 1990 CQR program. Also, there was no mechanism to prioritize cases
based on the magnitude of the error. As a result, regional offices wound
up

8Congressional Research Service, House Apportionment: Could Census
Corrections Shift a House Seat?, RS21638 (Washington, D.C.: Oct. 8, 2003).

expending considerable resources on CQR cases that only affected a handful
of dwellings.

The draft evaluation also found that the two software applications the
Bureau chose to administer and track CQR cases did not appear to be up to
the task. Lost cases and documentation, poor integration with other
applications, and the inability to produce reports were among the issues
the evaluation cited.

More generally, the integration and coordination issues that affected CQR
are not unique to that program; to the contrary, our past reports have
found that other components of the 2000 Census were not well planned,
which unnecessarily increased the cost and risk of the entire
enumeration.9 The need for better strategic planning has been a consistent
theme in many of our past recommendations to improve the Bureau's approach
to counting the nation's population and represents a significant
management challenge that the Bureau will need to address as it looks
toward 2010.

The Bureau Is Making Count Correction Plans for 2010

The Bureau is beginning to develop plans for Full Count Review and CQR for
the 2010 Census. As it does so, it will be important for it to develop an
initiative or consolidated program that corrects both systemic and
individual issues, and does so prior to the release of apportionment and
redistricting data. Granted, this effort will be no simple task given the
relatively short time between the closure of the local census offices and
the need to release the public law data within the legally required time
frames.

Still, there are steps the Bureau can explore to methodically check the
data for nationwide systemic errors, obtain local input, and investigate
any discrepancies, and do so in an expeditious manner. One approach might
be to consolidate and leverage CQR, Full Count Review, and certain other
Bureau programs.

Indeed, under the Full Count Review program, the Bureau obtained local
input by contracting out some of the work to members of the Federal-State
Cooperative Program for Population Estimates (FSCPE), an organization

9See for example, GAO, 2000 Census: Lessons Learned for Planning a More
Cost-Effective
2010 Census, GAO-03-40 (Washington, D.C.: Oct. 31, 2002), 14 - 17, and
GAO, 2010 Census:
Cost and Design Issues Need to Be Addressed Soon, GAO-04-37 (Washington,
D.C.:
Jan. 15, 2004), 25 - 31.

composed of state demographers that has worked with the Bureau since 1973
to ensure accurate population counts. The Bureau worked with FSCPE, in
part, because it lacked sufficient staff to complete the review on its
own, but also because the Bureau believed that the members' knowledge of
the demographic characteristics of their states could help the Bureau
examine data files and products, including public law data. FSCPE members
reviewed data for 39 states and Puerto Rico; Bureau employees reviewed
data for the remaining states and the District of Columbia without FSCPE
representation in Full Count Review. Both sets of analysts checked the
data for their overall reasonableness, as well as for their consistency
with historical and demographic data, and other census data products.

Bureau staff from its regional offices reviewed the data as well. They
focused on identifying inconsistent demographic characteristics and did
not necessarily concentrate on any one particular state or locality. Thus,
the Bureau obtained local input that focused on individual states and
smaller jurisdictions, and also performed its own, broader review.

Verifying any data discrepancies could be accomplished by beginning the
count correction effort as local census offices complete nonresponse
follow-up, when enumerators are still available to investigate issues. In
fact, the Bureau is already planning to do this to some degree in 2010
under another operation called Coverage Improvement Follow-up (CIFU),
where the Bureau is to call or visit housing units that have been
designated as vacant or nonexistent but not confirmed as such by a second
source. In the 2000 Census, CIFU began June 26, 2000, and ended on August
23, 2000. During that time, enumerators contacted 8.9 million housing
units and counted 5.3 million people, according to the Bureau.

The Bureau could explore adding the count correction workload to
enumerators' CIFU assignments, which would enable the agency to reconcile
possible data errors, as well as add any housing units and group quarters
the Bureau missed during the initial enumeration (As noted in the
background section, CQR could not add any residences that existed on
Census Day but the Bureau had failed to count.).

Further, the Bureau could help automate the count correction process by
using computers to flag any data that exceed any predetermined tolerances.
The Bureau could also develop a system to prioritize count correction
issues to help manage its verification workload.

Importantly, to the extent the Bureau reviews and corrects census counts
prior to the release of the public law data, the Bureau might not need
separate Full Count Review and CQR programs; a consolidated effort might
be more cost effective. At a minimum, to the extent a separate CQR program
is needed, it may not have to be as large or last as long because
presumably the earlier program would have caught the bulk of the problems.

Regardless, given the possibility that similar data errors might again
occur during the 2010 Census, exploring options for resolving them prior
to the release of public law data would be a sound investment.
Reapportionment and redistricting data would be more accurate; the
Bureau's credibility would be enhanced; and the need for a large-scale
count correction program along the lines of CQR could be reduced or
eliminated.

Conclusions	The CQR program played an important role in improving the
quality of data from the 2000 Census. Although the net changes in housing
and population counts from the program were small on a national scale, in
a number of instances, they were substantial at the local level, and could
affect various revenue sharing formulas and other programs that use
decennial census data.

Because the program functioned as a safety net-a final opportunity to
catch and correct mistakes that occurred along the chain of events that
led to, and extended beyond Census Day 2000-the results shed light on the
performance and limitations of certain upstream census operations, and
areas where the Bureau should focus its efforts as its plans unfold for
2010. In this regard, the following is clear: although the Bureau puts
forth tremendous effort to ensure a complete and accurate census, its
numerous procedures and quality assurance operations will be challenged to
stay ahead of the increasing difficulties associated with enumerating a
population that is growing larger, more diverse, and increasingly hard to
locate and count.

The timing of any count correction effort will also be critical. Indeed,
we are concerned that key decisions using data from the 2000 Census
employed figures that, for a number of jurisdictions, were later found to
be flawed. As a result, it will be important for the Bureau to consider
developing a count correction initiative that can complete its work in
time to correct the public law data before that information is released.

Moreover, beyond the inherent demographic obstacles to a successful
census, the results of our CQR review echo several of our past reports on
other aspects of the census, which note that some of the Bureau's
difficulties stem from a lack of adequate strategic planning and other
management challenges. Ultimately, the success of the 2010 Census will
hinge on the extent to which senior Bureau leadership resolves these
challenges.

With this in mind, resolute action is needed across three fronts. First,
it will be important for the Bureau to ensure, via thorough field testing,
that its planned changes to its address list development procedures help
resolve the geocoding and other operational problems revealed by CQR.
Second, it will be important for the Bureau to improve its count
correction efforts by designing a program that can systematically and
consistently review the public law data and make any corrections prior to
the release of those figures. Third, it will be important for the Bureau
to address persistent strategic planning challenges.

Recommendations for 	To help ensure the nation has the best possible data
for purposes of apportionment, redistricting, and other uses of census
data, we

Executive Action	recommend that the Secretary of Commerce direct the
Bureau to improve its count correction efforts for the 2010 Census by
taking such actions as:

1.

2.

3.

Thoroughly testing improvements to the Bureau's group quarters and other
address list development activities to help ensure the Bureau has resolved
geocoding and other problems with its master address file.

Consolidating Full Count Review and CQR into a single program that
systematically reviews and corrects any errors in the public law data
prior to their release.

Expediting count correction efforts by initiating data reviews toward the
end of nonresponse follow-up, when the Bureau starts getting complete data
for geographic entities, and enumerators are available to help investigate
any discrepancies. As part of this effort, the Bureau should consider
using computers to systematically search for possible errors nationwide by
checking data at the appropriate level of geography to ensure population,
housing unit, and group quarter counts, as well as demographic
characteristics, appear reasonable and are consistent with population
estimates. Those areas that are outside of predetermined tolerances should
be flagged for further review. The

Bureau should also pay special attention to ensure group quarters are
properly geocoded and counted.

4.	Prioritizing the investigation of errors based on the magnitude of the
suspected error or similar triaging formula.

5.	Ensuring that instructions on the Bureau's Web site make it clear that
updated information exists and that users can readily access this
information.

6.	Improving the Bureau's quality assurance procedures to help ensure
there are no mistakes in the data the Bureau posts on its Web site.

7.	Enhancing the training and guidance provided to regional offices to
help ensure they share the same understanding of their roles and
responsibilities and will implement the program consistently.

8.	Addressing persistent strategic management challenges, in part, through
early testing to help ensure information systems, training, and other
activities are fully integrated.

Agency Comments and Our Evaluation

The Acting Deputy Secretary of Commerce provided written comments on a
draft of this report on May 20, 2005, which are reprinted in appendix III.
Commerce stated that "the report provides a good overview of program
results and makes several useful observations and recommendations," and
specifically agreed with our finding that the process for conducting
internal reviews was not consistently implemented. More generally,
however, Commerce believes the shortcomings we describe reflect "a
fundamental misunderstanding of the goals of the CQR program," and noted
that our observations and recommendations indicate we believe that CQR
should have been designed to correct the public law data before they were
released during the 2000 Census.

Our concern over the CQR program centers on the way it was implemented in
2001, rather than the fact that the Bureau did not design the program to
correct the apportionment and redistricting numbers. We agree with
Commerce that this was not the intent of CQR and, as Commerce notes, we
acknowledge this in our report. At the same time, based on the lessons
learned from the 2000 Census, enumeration errors are almost inevitable.
Thus, our recommendations focus on the future, and specifically, the
importance of developing mechanisms for the 2010 Census to review and

correct errors in the public law data to the greatest extent possible
before they are released. We have clarified the report to better reflect
GAO's position.

Commerce specifically addressed two of our eight recommendations,
disagreeing with both of them. With respect to our recommendation to
consolidate Full Count Review and CQR into a single program for the 2010
Census, Commerce noted that preliminary counts at the census tract or
block level are needed to conduct an effective CQR program, and that
information is not available until close to the deadline for releasing the
apportionment data. Commerce maintains there would be little opportunity
for local entities to review the counts and document potential problems
and even less time for the Bureau to conduct the necessary research and
field work.

Our report recognizes that it would be a challenge for the Bureau to
review and correct census figures and still release the public law data by
the legally required time frames. Still, as we note in our report, we
believe the Bureau could expedite the process by taking such steps as (1)
using computers to check census data for their overall reasonableness and
flagging areas that exceed predetermined tolerances; (2) focusing on known
trouble spots such as group quarters; and (3) beginning the review process
earlier, such as, when local census offices complete their nonresponse
follow-up efforts.

Moreover, as we state in the report, during the 2000 Census, the Bureau
already had programs in place that obtained local input on the census
numbers before the release of the public law data (Full Count Review), and
conducted extensive field operations to investigate certain discrepancies
(Coverage Improvement Follow-up). We believe that it will be important for
the Bureau to not simply replicate these programs for the 2010 Census or
make incremental improvements, but to see whether these programs could be
better leveraged and be more strategically employed to improve the
accuracy of the apportionment and redistricting data.

The other recommendation that Commerce specifically addressed was our call
for the Bureau in 2010 to prioritize the investigation of errors based on
the magnitude of the suspected problem. Commerce maintains that the
Bureau's policy in 2000 was to handle cases in the order they were
received from local jurisdictions, and asserts this was a fair and
reasonable practice. While this practice is not unreasonable, we continue
to believe that it would be more cost-effective for the Bureau to give
priority to those cases

where it could achieve a greater return on its investment in resources
(especially given our findings involving group quarters such as prisons
and college dormitories that affected relatively large population
clusters). Our recommendation echoes the Bureau's draft evaluation of the
CQR program, which noted that regional offices expended considerable
resources on CQR cases that affected only a handful of dwellings.
Moreover, as we state in our report, prioritizing the Bureau's workload
could help expedite the count correction process.

Commerce's comments also included some technical corrections and
suggestions where greater clarity was needed. We revised the report as
appropriate.

We will send copies of this report to the Chairman of the House Committee
on Government Reform, the Secretary of Commerce, and the Director of the
U.S. Census Bureau. Copies will be made available to others on request.
This report will also be available at no charge on GAO's home page at
http://www.gao.gov.

If you or your staff have any questions about this report, please contact
me at (202) 512-6806 or [email protected]. Contact points for our Office
of Congressional Relations and Public Affairs may be found on the last
page of this report. GAO staff who made major contributions to this report
are listed in appendix IV.

Orice M. Williams Director Strategic Issues

Appendix I

Change in State Populations As a Result of Count Question Resolution
Program

                        Page 31 GAO-05-463 Data Quality
         2000                    Total                                                                                                                                                                                                                                                                        District                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           
State   Census   CQR total  population U.S.     281,    281, 2,697 Alabama 4,447,100 4,447,351  Alaska 626,932 626,931  Arizona 5,130,632 5,130,632  Arkansas 2,673,400 2,673,400  California 33,871,648 33,871,653  Colorado 4,301,261 4,302,015  Connecticut 3,405,565 3,405,602  Delaware 783,600 783,600     of    572,059 572,059  Florida 15,982,378 15,982,824  Georgia 8,186,453 8,186,816  Hawaii 1,211,537 1,211,537  Idaho 1,293,953 1,293,956  Illinois 12,419,293 12,419,647  Indiana 6,080,485 6,080,517  Iowa 2,926,324 2,926,382 58 Kansas 2,688,418 2,688,824 406 Kentucky 4,041,769 4,042,285 516 Louisiana 4,468,976 4,468,958 -18 Maine 1,274,923 1,274,923 0 Maryland 5,296,486 5,296,507 21 Massachusetts 6,349,097 6,349,105 8 Michigan 9,938,444 9,938,480 36 Minnesota 4,919,479 4,919,492 13 Mississippi 2,844,658 2,844,656 -2 Missouri 5,595,211 5,596,683 1,472 Montana 902,195 902,195 0 Nebraska 1,711,263 1,711,265 2 Nevada 1,998,257 1,998,257 0    New    1,235,786 1,235,786 0  New   8,414,350 8,414,347 -3  New   1,819,046 1,819,046 0 New  18,976,457 18,976,821 364                              
        total    population     change Total 421,906 424,603                                                                                                                                                                                                                                                  Columbia                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             Hampshire                       Jersey                        Mexico                       York                            North                       
      population                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             Carolina 8,049,313 8,046,485 -2,828

Appendix I
Change in State Populations As a Result of
Count Question Resolution Program

                         (Continued From Previous Page)

       State         2000 Census total population CQR Total population change 
                                     total population 
    North Dakota                      642,200 642,200                       0 
        Ohio                    11,353,140 11,353,145 
      Oklahoma                    3,450,654 3,450,652 
       Oregon                     3,421,399 3,421,436 
    Pennsylvania                12,281,054 12,281,054 
    Rhode Island                  1,048,319 1,048,319 
South Carolina                 4,012,012 4,011,816 
    South Dakota                      754,844 754,844 
     Tennessee                    5,689,283 5,689,267 
       Texas                    20,851,820 20,851,790 
        Utah                      2,233,169 2,233,198 
      Vermont                         608,827 608,827 
      Virginia                    7,078,515 7,079,030 
     Washington                   5,894,121 5,894,141 
West Virginia                  1,808,344 1,808,350 
     Wisconsin                    5,363,675 5,363,715 
      Wyoming                         493,782 493,782                       0 
    Puerto Rico                   3,808,610 3,808,603                      -7 
                  Source: GAO analysis of U.S. Census 
                             Bureau data.             

Appendix II

Human Error and Other Factors Contributed to University of North Carolina
Counting Errors

The duplicate counting of nearly 2,700 students at the University of North
Carolina (UNC) at Chapel Hill during the 2000 Census resulted from a
combination of factors. The incident is interesting because it shows how
the various safety nets the Bureau has built to ensure an accurate count
can be undermined by human error, the limitations of census-taking
operations, and other events that in some cases occur years before Census
Day (April 1, 2000).

The duplicate count was discovered after CQR began when the director of
the Charlotte regional office (a UNC graduate), asked one of her
geographers (also a UNC graduate), to see whether the UNC dormitories were
counted in their correct locations. According to the geographer, the
director's curiosity was aroused after the CQR program found problems with
the geocoding of dormitories at other schools in the Charlotte region. The
geographer told us he initiated an internal CQR case in the summer of 2001
after discovering that two UNC dormitories were geocoded to the wrong
census block. Upon further research, where he reviewed information from
the census address file and the UNC Web site, the geographer concluded
that, in addition to the geocoding error, a large number of dormitories
and their occupants were counted in error.

Ultimately, by matching census records, the Bureau determined that 1,583
dormitory rooms in 26 buildings-and the 2,696 students who had resided in
them-were included twice in the 2000 Census. On the basis of our
interviews with Bureau staff and review of pertinent documents, the
following sequence of events led to these erroneous enumerations:

The Bureau divides the places where people live into two broad categories:
group quarters, which include prisons, dormitories, and group homes; and
housing units, which consist of single family homes, apartments, and
mobile homes. During the 2000 Census, the Bureau had distinct procedures
for building its group quarters and housing unit address lists and
enumerating their residents. For example, the Bureau typically enumerates
college dormitories by working with schools to distribute census
questionnaires to students. Conversely, the Bureau enumerates residents of
housing units by delivering questionnaires directly to them through the
mail. In the UNC situation, the 26 UNC dormitories were listed correctly
in the Bureau's group quarters database and incorrectly in the Bureau's
housing unit database.

Concerned there could be systemic issues with the Bureau's address list,
staff at the Bureau's headquarters investigated the source of the problem

Appendix II
Human Error and Other Factors Contributed
to University of North Carolina Counting
Errors

following the initial discovery by Charlotte employees. The headquarters
review found that the dormitories were improperly included in the U.S.
Postal Service's address file, which it initially shared with the Bureau
in November 1997 and continued to update through early 2000. The Bureau
uses this database to help build its housing unit address list.
Specifically, the Bureau discovered that the data field that normally
contains a street address erroneously contained a unit number and the name
of a UNC dormitory. The Bureau had no explanation for how the dormitory
names got into the U.S. Postal Service's address file.

Other procedures designed to verify census addresses produced conflicting
results, compounding the problem. One procedure in 1998 mistakenly
confirmed the dormitories as housing units, while another procedure-
called block canvassing-correctly flagged the addresses for deletion from
the Bureau's housing unit address list. However, under the Bureau's
protocols, to ensure an address was not improperly removed from the
census, an address had to be flagged twice to be deleted. During
nonresponse follow-up in 2000, where enumerators visited housing units
that failed to send back the questionnaires that were mailed to them, the
Bureau had a third opportunity to uncover the error. Because the
enumerators involved in this operation provided inconsistent information,
the Bureau ultimately did not delete any housing units included in the
initial census.

As part of the CQR case analysis, staff in the Bureau's Decennial
Statistical Studies Division checked the Bureau's address file for any
records that contained the word "dorm" in the address field to determine
whether a similar duplication occurred at other schools. This would have
picked up the word "dormitory" and its variants. On the basis of this
search, the Bureau concluded that a similar issue was not problematic
elsewhere in the country.

Appendix III

Comments from the Department of Commerce

Appendix III
Comments from the Department of
Commerce

Appendix III
Comments from the Department of
Commerce

Appendix III
Comments from the Department of
Commerce

Appendix III
Comments from the Department of
Commerce

Appendix IV

                     GAO Contact and Staff Acknowledgments

GAO Contact Orice Williams, (202) 512-6806

Acknowledgments	In addition to the contact named above, Robert Goldenkoff,
Keith Steck, Timothy Wexler, Robert Parker, Michael Volpe, Andrea Levine,
and Elena Lipson made key contributions to this report.

GAO's Mission	The Government Accountability Office, the audit, evaluation
and investigative arm of Congress, exists to support Congress in meeting
its constitutional responsibilities and to help improve the performance
and accountability of the federal government for the American people. GAO
examines the use of public funds; evaluates federal programs and policies;
and provides analyses, recommendations, and other assistance to help
Congress make informed oversight, policy, and funding decisions. GAO's
commitment to good government is reflected in its core values of
accountability, integrity, and reliability.

Obtaining Copies of The fastest and easiest way to obtain copies of GAO
documents at no cost

is through GAO's Web site (www.gao.gov). Each weekday, GAO postsGAO
Reports and newly released reports, testimony, and correspondence on its
Web site. To Testimony have GAO e-mail you a list of newly posted products
every afternoon, go to

www.gao.gov and select "Subscribe to Updates."

Order by Mail or Phone	The first copy of each printed report is free.
Additional copies are $2 each. A check or money order should be made out
to the Superintendent of Documents. GAO also accepts VISA and Mastercard.
Orders for 100 or more copies mailed to a single address are discounted 25
percent. Orders should be sent to:

U.S. Government Accountability Office 441 G Street NW, Room LM Washington,
D.C. 20548

To order by Phone:	Voice: (202) 512-6000 TDD: (202) 512-2537 Fax: (202)
512-6061

To Report Fraud, Contact:
Waste, and Abuse in Web site: www.gao.gov/fraudnet/fraudnet.htm

E-mail: [email protected] Programs Automated answering system: (800)
424-5454 or (202) 512-7470

Congressional	Gloria Jarmon, Managing Director, [email protected] (202)
512-4400 U.S. Government Accountability Office, 441 G Street NW, Room 7125

Relations Washington, D.C. 20548

Public Affairs	Paul Anderson, Managing Director, [email protected] (202)
512-4800 U.S. Government Accountability Office, 441 G Street NW, Room 7149
Washington, D.C. 20548
*** End of document. ***